CN116802620A - Apparatus and method for remote direct memory access - Google Patents

Abstract

The present invention relates to data communication between devices such as requesting devices and responding devices via RDMA operations. The requesting device and the responding device each include memory, a controller, and a communications interface. The controller of the requesting device is associated with a memory area of the responding device, the memory area including one or more memory blocks. Furthermore, the responsive device associates the memory area in its memory, determines a memory status of the memory area, and transmits a notification regarding the memory status related to the memory area through the communication interface. In addition, based on the memory status, the requesting device determines whether the memory block to be accessed is ready in the memory of the responding device, and then accordingly transmits a request to the responding device through the communication interface, the request Relevant to the memory block to be accessed.

Description

Apparatus and method for remote direct memory access

Technical field

The present invention relates generally to data transfer; more particularly, to apparatus and methods for remote direct memory access.

Background technique

In recent years, the high-performance computing (HPC) field has leveraged high-performance and low-latency networks such as InfiniBand on existing IP infrastructure and customized lossless technologies. These interconnects provide advanced network capabilities such as remote direct memory access (RDMA) to achieve high throughput and low latency as well as low CPU utilization. RDMA is being used in high-performance computing (HPC), data centers, Web 2.0, big data, cloud, storage, financial services and insurance (FSI), medical services, federal and other fields. However, high-speed data communication in networks of multiple computing devices increases concerns about data reliability and latency.

Generally, remote direct memory access (RDMA) involves direct memory access from the memory of one computing device to the memory of another computing device. Generally, RDMA technology does not require an operating system (OS) of any computing device (i.e., a traditional sending device and a traditional receiving device), can perform high-throughput, low-latency data transmission on the network, and achieves high performance. Especially useful in critical deployments. There are two common RDMA technologies, InfiniBand and Internet Engineering Task Force (IETF). The IETF protocol is also called the Internet wide Area RDMA protocol (iWARP). Specifically, InfiniBand RDMA comes in two variants that allow it to run over Ethernet networks (RoCE) and over UDP/IP networks (RoCEv2). For example, in an exemplary implementation scenario of a legacy RDMA system, such as during an RDMA read or write operation, data (or packets) are transferred directly from source memory (eg, of a legacy requesting device) to (eg, a legacy requesting device) using an RDMA-capable network adapter. , traditional response device) target memory. Specifically, it does not involve the operating system of any one of the central processing unit (CPU) (or processor) or the interactive devices (ie, traditional sending devices and traditional receiving devices). However, unlike the requester (e.g., traditional requesting device) that accesses the memory, which has knowledge about the packet (QP context, operational information) when the packet is generated, the responder (e.g., traditional responding device) receives the packet from the network. to receive information related to a specific operation. As a result, the process gains considerable processing time and increases latency during data communication. Additionally, performing these operations at the line speeds of modern RNICs (25–400Gbps) becomes extremely challenging. Therefore, in traditional systems, retrieving information from host memory takes time and often requires several data retrieval iterations, resulting in increased latency and reduced performance.

Currently, host memory is very expensive (estimated to be up to 40% of the total host cost) and resources are limited. The availability of this limited resource is enhanced by the need to reserve memory for DMA operations, making reserved or physical memory available to the RDMA NIC as needed. Traditionally, there are two ways to overcome the above problem, either by pinning memory or by employing demand paging. However, pinned memory, which accounts for a large portion of available physical memory, effectively reduces the physical memory available on the host device because it cannot be used for other purposes during periods of inactivity (i.e., not used for DMA operations). Therefore, no successful trade-off is established between DMA operations and resources (i.e., memory or RAM). Additionally, on-demand paging (ODP) requires that missing page mappings be handled by dynamically mapping pages to DMA destinations during execution without direct intervention from the user application. While ODP is possible, it may result in performance degradation such as reduced speed or execution time, reduced inbound network traffic, or rate throttling, resulting in significant negative performance impact. Typically, prefetching data on the responder before DMA operations can reduce performance overhead by sending a prefetch hint request to the responder before each DMA operation, thereby mitigating the performance impact of demand paging. However, the prefetch hint request itself for every or every few DMA operations creates bandwidth overhead on the network. Additionally, redundant prefetch hint requests can delay the processing of useful prefetch hint requests when the page is already mapped.

Therefore, in accordance with the above discussion, there is a need to overcome the above-mentioned shortcomings associated with conventional requesting devices, conventional responding devices and conventional methods of conventional remote direct memory access (RDMA) technology for data communication.

Contents of the invention

The present invention seeks to provide requesting devices, responding devices, methods and systems for remote direct memory access (RDMA). The present invention seeks to solve the existing inefficient and unreliable data communication problems with the help of traditional equipment, methods and traditional remote direct memory access systems. It is an object of the present invention to provide a solution that at least partially overcomes the problems encountered in the prior art and to provide improved devices, methods and systems for efficient (eg, reduced latency) and reliable data communication for RDMA operations. The object of the invention is achieved by the technical solutions provided in the appended independent claims. Advantageous implementations of the invention are further defined in the dependent claims.

The object of the invention is achieved by the technical solutions provided in the appended independent claims. Advantageous implementations of the invention are further defined in the dependent claims.

In one aspect, the invention provides a request device including a memory, a controller and a communication interface. The controller is used to associate with the memory area of the responding device, which includes one or more memory blocks. In addition, the controller is configured to receive notifications about the memory status related to the memory area from the responding device through the communication interface; determine whether the memory block to be accessed is ready in the memory of the responding device based on the memory status, and transmit the request to the responding device through the communication interface , the request is related to the memory block to be accessed.

The requesting device of the present invention is able to keep track of the memory status of a memory area or one or more memory blocks, and allows the responding device to optimize various aspects with the help of memory prefetching, prefetch hints, etc. These operations are used to enable responding devices to start loading memory blocks or memory regions in advance (if necessary), or to prepare to accept incoming packets in an efficient manner. Additionally, faster and more efficient operations can be achieved by transferring multiple notifications related to multiple transactions (or operations) at once by swapping memory areas when needed. Responding devices report updates as notifications so that requesting devices have a higher chance of being served without page faults.

In another implementation form, the controller is further configured to receive an indication of a memory block size that is less than or equal to the size of the memory region, wherein notifications about memory status relate to one or more memory blocks.

The indication of the memory block size enables optimization of the occurrence and control overhead of page faults during operation and ensures efficient data communication by allowing the requesting device to reconfigure or change the granularity or structure of the request based on the received indication of the memory block size.

In another implementation form, the controller is further configured to transmit the indicated memory block size to the responding device prior to receiving the indication of the memory block size to indicate to the responding device the memory state granularity supported by the requesting device.

The indicated memory block size transferred allows the responding device to reconfigure or change the granularity of the memory blocks in the responding device based on the indicated size transferred by the requesting device to optimize the occurrence and control overhead of page faults during operation and to ensure efficient memory status reporting communication .

In yet another implementation form, the controller is further configured to initiate a negotiation process with the responding device for a memory block size based on one or more of bandwidth criteria, responding device capabilities, and requesting device capabilities.

The negotiation process allows bandwidth to be enhanced by optimizing or changing the granularity on the responding device side.

In another implementation, the controller is also configured to receive an indication of an updated memory block size.

The updated indication of memory block size enables the requesting device to make informed decisions before transmitting the request to speed up data communication and avoid possible page faults and errors.

In another implementation form, the updated memory block size is based on one or more of the number of memory actions, bandwidth usage, responding device capabilities, and requesting device capabilities.

Updated memory block sizes based on the number of memory actions, bandwidth, and device capabilities allow factors to be optimized by selecting or updating the size or granularity of the memory blocks.

In one implementation, the controller is also used to store the memory state of the memory block.

The requesting device keeps track (or bookkeeping) of memory state to enable faster data communication and minimize page faults at the responding device.

In one implementation form, the controller is further configured to group multiple memory blocks into memory block groups, determine the memory group status of the memory block group based on the memory status of the multiple memory blocks, and store the memory group status.

Chunking or grouping of memory blocks enables the requesting device to clearly classify and track the memory bank status of multiple memory blocks for more efficient data communication. Determining the memory state of multiple memory blocks collectively rather than individually reduces the number of notifications sent over the network, thus reducing or saving database size in the requesting device and allowing for greater efficiency memory allocation.

In another implementation form, wherein the memory group status includes a counter, the controller is configured to: if the received memory status indicates that a memory block included in a plurality of memory blocks of the memory block group has changed its memory status, adjust the counter .

In another implementation form, wherein the memory group status includes a memory flag, the controller is configured to set the flag if the notification indicating the memory status indicates that a memory block of the memory block group is not currently in the memory of the responding device.

Counters and flags allow keeping track or bookkeeping of the memory status of a memory block or group of memory blocks to ensure efficient memory status reporting communication.

In one implementation, notifications about memory status indicate changes to one or more memory blocks.

Updated notifications enable requesting devices to configure requests based on changes in memory status to avoid page faults or errors and enable query-free operations.

In one implementation form, notifications about memory status indicate the current memory status of one or more memory blocks.

Notification about the current memory state enables the requesting device to configure requests based on the latest or current state of a memory region or memory block to avoid any or some page faults.

In one implementation form, the controller is further configured to: if it is determined that the memory block to be accessed is not ready in the memory of the responding device, send a message including a prefetch hint request indicating to the responding device that it is to be accessed. memory block.

In one implementation form, the notification regarding the memory status indicates that the responding device is incompatible, wherein the controller is further configured to: adjust memory requests for one or more memory blocks accordingly; and not send a message including a prefetch hint request.

This notification enables the requesting device to save computing time by indicating the deficiency at an early stage, so that messages including prefetch hint requests are not sent and further resources are not spent.

In another aspect, the present invention provides a method for a requesting device including a memory, a controller, and a communication interface. The method includes: associating with a memory area of the responding device, the memory area including one or more memory blocks; receiving notification about memory status from the responding device through the communication interface; based on the memory status Determine whether the memory block to be accessed is ready in the memory of the responding device; transmit a request to the responding device through the communication interface, the request being related to the memory block to be accessed.

The method of this aspect realizes all the advantages and effects of the requesting device of the invention.

In yet another aspect, the present invention provides a computer-readable medium carrying computer instructions that, when loaded into and executed by a controller of a requesting device, enable the requesting device to implement the method.

Computer-readable media (specifically, non-transitory computer-readable media) carrying computer instructions realize all the advantages and effects of the requested device or the method.

In one aspect, the invention provides a response device including a memory, a controller and a communication interface. The controller is configured to: associate a memory area with a requesting device, the memory area including one or more memory blocks; determine a memory state associated with the memory block; and transmit a notification regarding the memory state to the requesting device through the communication interface.

Since the memory state is associated with the request, the controller of the responding device determines the memory state with respect to the memory region and responds accordingly to allow the memory region to be reconfigured on either end for faster data communication. Beneficially, these notifications transmitted by the responding device greatly reduce the latency of servicing the RDMA message itself when it arrives at the responding device and make data communications more efficient.

In one implementation, the controller is further configured to transmit an indication of a memory block size that is smaller than the size of the memory region, wherein the notification of the memory status relates to one or more memory blocks.

Tracking the memory state of one or more memory blocks rather than the memory state of an entire memory region enables requests to be adjusted to achieve higher bandwidth, lower latency, and therefore faster, more efficient data communication.

In another implementation form, the controller is further configured to receive the indicated memory block size from the requesting device prior to transmitting the indication of the memory block size.

In one implementation form, the controller is further configured to initiate a negotiation process with the requesting device for a memory block size based on one or more of bandwidth criteria, responding device capabilities, and requesting device capabilities.

In one implementation, the controller is also configured to determine a memory state of one of one or more memory blocks.

The requesting device keeps track (or bookkeeping) of the memory state of the block or blocks of memory that the requesting device is interested in, saving time for faster data communication and minimizing page faults.

In one implementation form, the controller is further configured to determine a memory state of the memory block based on a change in the memory state of the memory block.

This allows the controller to provide the latest or current and accurate memory status of the memory block for efficient and reliable data transfer without possible delays.

In one implementation form, the controller is further configured to determine a memory state of the memory block based on determining the memory location of the memory block.

In another implementation form, the controller is further configured to determine the memory status of one or more memory blocks within a time period, and when the time period expires, transmit a memory status indicating the memory status of the one or more memory blocks. notify.

When the memory state changes, a notification is transmitted after a period of time. If a notification is already scheduled, append the status change to the exit notification.

In yet another implementation form, the controller is further configured to determine a memory status of one or more memory blocks; determine a number of memory actions; determine that the number of memory actions exceeds a threshold number; and then transmit a memory indicating that the one or more memory blocks status notification.

Determination of memory status and number of memory actions allows appending memory status based on threshold number of pages swapped in/out, time period, or maximum notifications reported per message to optimize data communication.

In one implementation form, the controller is further configured to receive a prefetch hint request from the requesting device through the communication interface, the prefetch hint request indicates one or more memory blocks, and load the indicated one or more memory blocks into the memory. middle.

Receives a prefetch hint request to indicate the requesting device's requirements or to enable complete mapping of a memory region or memory block, thus reducing or eliminating wait cycles in the event that an unmapped memory region exists during an RDMA operation.

In another aspect, the present invention provides a method for a response device, the response device includes a memory, a controller and a communication interface, the method includes associating a memory area with the requesting device; determining whether the memory area is relevant to the requesting device. memory status, and transmits a notification about the memory status to the requesting device through the communication interface.

The method of this aspect realizes all the advantages and effects of the receiving device of the invention.

In one implementation form, the present invention provides a computer-readable medium carrying computer instructions that, when loaded into and executed by a controller of a response device, enable the response device to implement the described method.

A computer-readable medium carrying computer instructions realizes all the advantages and effects of a responsive device or method.

It should be understood that all the above embodiments can be combined together.

It should be noted that all devices, components, circuits, units and modules described in this application can be implemented in software components or hardware components or any type of combination thereof. The steps performed by various entities described in this application and the functions described to be performed by various entities are intended to mean that each entity is used to perform the various steps and functions. Even if in the following description of specific embodiments, a specific function or step to be performed by an external entity is not reflected in the description of the specific detailed elements of the entity that performs the specific step or function, it should be clear to the skilled person that these methods and functions may be implemented accordingly. implemented in software or hardware components, or in any combination of such components. It will be appreciated that the features of the invention are susceptible to being combined in various combinations without departing from the scope of the invention as defined by the appended claims.

Other aspects, advantages, features and objects of the invention will become apparent from the accompanying drawings and the detailed description of illustrative implementations, taken in conjunction with the following appended claims.

Description of the drawings

The foregoing summary and the following detailed description of illustrative embodiments may be better understood when read in conjunction with the accompanying drawings. For the purpose of illustrating the invention, exemplary structures of the invention are shown in the drawings. However, this invention is not limited to the specific methods and instrumentalities disclosed herein. Furthermore, those skilled in the art will appreciate that the drawings are not drawn to scale. Where possible, identical components are designated by the same numbers.

The embodiments of the present invention are described below by way of example only and in combination with the following drawings.

Figure 1 is a block diagram of a requesting device that can be used to perform remote direct memory access (RDMA) operations according to an embodiment of the present invention;

Figure 2 is a flow chart of a method for requesting a device to perform an RDMA operation according to an embodiment of the present invention;

Figure 3 is a block diagram of a response device that can be used to perform RDMA operations according to an embodiment of the present invention;

Figure 4 is a flow chart of a method for responding to a device performing an RDMA operation according to an embodiment of the present invention;

Figure 5 is a schematic diagram depicting a system for performing RDMA operations according to an embodiment of the present invention; and

6-7 are schematic diagrams depicting exemplary scenarios for implementation of an RDMA system according to various embodiments of the present invention.

In the drawings, underlined numerals are used to represent the item in which the underlined numeral is located or an item adjacent to the underlined numeral. An ununderlined number is associated with the item identified by the line linking the ununderlined number to the item. When a number is not underlined but has an associated arrow, the ununderlined number is used to identify the general item to which the arrow points.

Detailed ways

The following detailed description illustrates embodiments of the invention and its implementation. Although a few embodiments of this invention have been disclosed, those skilled in the art will recognize that other embodiments for implementing or practicing the invention may be realized.

Referring now to FIG. 1 , shown is a block diagram of a requesting device 100 that may be used to perform remote direct memory access (RDMA) operations in accordance with an embodiment of the present invention. FIG. 1 generally illustrates various exemplary components of the requesting device 100. As shown, requesting device 100 includes memory 102, controller 104, and communication interface 106.

The requesting device 100 can potentially function as a sending device (or transmitter). Examples of requesting devices 100 may include, but are not limited to, communication channel adapters, servers, computing or computing devices in computer clusters (eg, massively parallel computer clusters), communications devices including portable or non-portable electronic devices, wireless modems, supercomputers or other RDMA based devices.

Memory 102 includes suitable logic, circuitry, and/or interfaces for storing instructions executable to control the operation of requesting device 100 . For example, memory 102 may store an operating system and/or other program products to operate requesting device 100 . The memory 102 may store data (transmitted in the form of data packets) for processing at the requesting device 100 . Examples of implementations of memory 102 may include, but are not limited to, electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), read only memory, ROM), hard disk drive (HDD), flash memory, secure digital (SD) card, solid-state drive (SSD) and/or CPU cache memory. Computer-readable storage media used to provide non-transitory memory may include, but are not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing.

The communication interface 106 of the requesting device 100 includes a medium (eg, a communication channel) over which the requesting device 100 communicates with the responding device. Examples of the communication interface 106 include, but are not limited to, communication channels in a computer cluster, local area communication channel (LAN), cellular communication channel, wireless sensor communication channel (WSN), cloud communication channel, vehicle A vehicle-to-communication channel (V2N) communication channel, a metropolitan area communication channel (MAN), and/or the Internet. In addition, the requesting device 100 in the communication channel environment is used to connect to the responding device according to various communication channel protocols supporting RDMA communication. Examples of such communication channel protocols, communication standards and technologies may include, but are not limited to, infiniband (IB), RDMA over converged ethernet (RoCE), Internet wide area RDMA protocol (iWARP) ) or modifications and variations thereof.

In the present invention, the controller 104 is an RDMA controller that includes suitable logic, circuitry, and/or interfaces for generating a prompt request that includes an indication of the type of operation. Controller 104 is a computing element for processing instructions to drive requesting device 100 . Examples of the controller 104 include, but are not limited to, a network interface controller, a microprocessor, a microcontroller, a complex instruction set computing (CISC) microprocessor, a reduced instruction set (RISC) microprocessor or very long instruction set (VLIW) microprocessor. The controller 104 may include a memory management unit (MMU) to allocate memory areas. Alternatively, one or more controllers, interactive devices (eg, requesting device 100, responding device) and elements similar to controller 104 may be arranged in various architectures for responding to and processing instructions driving requesting device 100.

The controller 104 is configured to associate with a memory area of the response device, where the memory area includes one or more memory blocks. Optionally, controller 104 transmits a request through communication interface 106 to associate a memory region with a responding device (similar to requesting device 100). The request is based on the request-response or request-reply method used for communication between different computing devices in the network. For example, a first computing device (eg, requesting device 100) sends a request to associate a block of data or memory, and a second computing device (eg, responding device) responds to the request. Specifically, the request is part of a message exchange pattern in which the requesting device 100 transmits a request to associate a memory area with a responding device, where the responding device receives and processes the request and ultimately returns the memory associated with the requested zone-related messages in response. It should be understood that association and memory registration can be implemented in other different ways without limiting the scope of the invention. A "memory region" refers to a region, also called a resource partition, extent, or memory context, which is a collection of allocated blocks that can be efficiently freed based on implementation. As used herein, a memory area is a portion of a larger memory such as memory 102. Typically, all blocks in a memory region are allocated in a single contiguous memory address range or multiple memory address ranges, similar to traditional stack frame allocation. Memory 102 includes one or more memory areas that may be allocated to different operations based on implementation. Typically, memory regions are created on a page basis, similar to the kernel size. Typically, memory areas can range from 1 byte, 2 bytes, 4 kilobytes (KB), 8KB, 64KB, 2 megabytes (MB), 8MB, 32MB, 1 gigabyte ( gigabyte, GB), 16GB, 128GB, etc. the entire memory 302 of the response device 300. It is worth noting that each memory "page" on the memory region is similar in size to the kernel size. The term "kernel" refers to the central component of an operating system that governs the operation of computing system software and hardware. Alternatively, the operations used by the kernel to manage memory and CPU time. The request includes information about memory areas and/or memory blocks to aid in the effectiveness of data communications. The memory area of memory 102 will be associated by the responding device with the memory of the responding device. The memory of the responding device is similar to the memory 102 of the requesting device 100 . Typically, a memory area consists of one or more memory blocks. The term "block" or "memory block" refers to a sequence of bytes or bits with a maximum length or block size, usually containing multiple records. Therefore structured data is considered to be chunked or called chunks. Typically, a memory area is divided into one or more parts, such as a first part i.e. 0-50% and a second part i.e. 50-100%. In one example, a 1GB memory area is divided into two memory blocks of 0.5GB each, and a trace for the first memory block will be triggered and messages for the first memory block will be triggered accordingly, and any changes on the memory block Both traces and messages associated with the second memory block will be triggered. It should be understood that this division is exemplary only and that any number or size of memory blocks may vary depending on the implementation without limiting the scope of the invention. Typically, the ratio of the size of a memory block to the size of a memory region is 1:1, but this ratio can be changed by the controller or computing device depending on the implementation.

The controller 104 is configured to receive notifications about memory status related to the memory region from the responding device through the communication interface 106 . "Notification" refers to the type of message sent from the responding device related to a memory area. The responding device, upon receiving a request from the controller 104 to associate a memory region (or page of memory), analyzes the memory to determine the memory status. "Memory State" means the state of an associated memory region, which may or may not be available or ready at the time of request, including one or more memory blocks. So basically, each of one or more memory blocks is screened or analyzed to determine their status. The collective state of each of the one or more blocks that form a memory region is called the memory state. Optionally, the memory state is related to a single memory block. Optionally, the memory state relates to the entire memory region. Typically, the memory status indicates whether the requested memory block or memory region is available, based on which the responding device transmits a notification to the requesting device 100 . The notification may include the memory status of each of the associated one or more memory blocks. Accordingly, the requesting device 100 or the controller 104 is configured to receive notifications from the responding device indicating memory status related to the memory region. Typically, a block of memory is either available or not. The available state of a memory block can be called the swap-in state, while the unavailable state of the memory can be called the swap-out state. The requesting device 100 is used to transmit a prompt message to enable the responding device to perform and prepare a swapping or swapping operation on one or more memory blocks according to the requirement of concurrently executing one or more operations. The term "swapping" refers to a memory management scheme in which any process or operation can be temporarily swapped from primary storage (e.g., the memory of a responding device) to secondary storage (e.g., backup drive, hard drive) to primary storage (e.g., random access memory) or RAM) can be used by other processes. Beneficially, memory swapping is employed to increase utilization of the memory 102 or memory of the responsive device. Specifically, if any memory block becomes unavailable for any reason, that memory block can be swapped out with the same memory block in secondary storage, and vice versa. Such a swap operation enables the responding device to enable the association of a memory region without incurring any further time delay due to the inherent need for further communication in the form of notifications regarding unavailable memory blocks. Beneficially, unavailable memory blocks are swapped out, and correlation can be performed in the event of any interruption.

In another embodiment, the controller 104 is further configured to receive an indication of a memory block size, where the memory block size is less than or equal to the size of the memory area. A memory block (or page) is a portion of the requested memory area of the responding device or the entire requested memory area. The controller 104 receives an indication of the memory block size, which is used by the responding device to generate notifications indicative of memory status. Depending on the implementation, the indication may include the size of one or more memory blocks. Based on the received indication, the requesting device 100 or the controller 104 may reconfigure the request or association accordingly based on the indication of the memory block size. For example, the controller 104 receives an indication that the memory block size is 40 kilobytes (KB), based on which the requesting device 100 is instructed to notify the granularity that further indicates the state of the memory for further operations. Typically, any requesting device may require memory blocks or memory regions of various sizes for efficient data communication. Different sizes of memory blocks or memory regions provide granularity to memory 102 and thus flexibility to the memory. Therefore, to avoid any discrepancies that might inhibit efficient data communication during operation, the responding device transmits the memory block size to assist the requesting device 100 in determining the accessibility of the associated memory block. A memory area is divided into equal logical "block sizes" or blocks of memory. The number of blocks may vary without limiting the scope of the invention. The number of blocks can range from one block for the entire memory area to a single block per page in the memory area. Additionally, notifications about memory status relate to one or more memory blocks. Typically, each of one or more memory blocks includes a separate memory state that indicates the status or availability of the memory block. Overall, each memory block helps determine the memory state of the memory region. Notably, the requesting device 100 is able to determine the notification granularity based on, but not limited to, the number of unmapped/fully mapped notifications, the number of page faults encountered, etc. and dictate the structure of the memory region depending on the implementation.

In another embodiment, the controller 104 is further configured to transmit the indicated memory block size to the responding device before receiving the indication of the memory block size. "Indicated size" means the indication provided to the responding device of the size of the memory block for which a memory status notification will be generated. The controller 104 transmits the indicated memory block size, where the indicated size is based on the memory block size in the memory 102 of the requesting device 100 . The term "indicated size" refers to a recommendation for the memory block size associated with the memory status of the responding device, which recommendation is subsequently communicated to the requesting device 100. In other words, the sizes indicated are recommendations for memory state granularity. It is worth noting that the associated memory area consists of one or more memory blocks, where each memory block has the same size. Optionally, the memory block size can be determined using the indicated size. It should be understood that notifications may be sent at any point in time depending on the implementation without limiting the scope of the invention. The responding device is used to notify the requesting device 100 of the selected granularity (eg, by the indicated size). In one example, when the number of unmapped pages for a block increases from zero to one or more, a notification is sent to the requesting device 100 indicating that the block may be unmapped. In another example, when the number of unmapped pages for a block decreases from one or more to zero, a notification is sent to the requesting device that the block may be mapped.

In one embodiment, notification regarding the indicated size may be transmitted upon association of the memory area. Alternatively, the notification can be transmitted after receiving the memory state of the associated memory region.

In yet another embodiment, the controller 104 is further configured to initiate a negotiation process with the responding device for a memory block size based on one or more of bandwidth standards, responding device capabilities, and requesting device capabilities. The controller 104 is used to initiate the negotiation process. "Negotiation process" refers to a simple protocol used to exchange and determine the size of memory blocks used by the requesting device 100 and the responding device. The bidirectional nature of the negotiation process allows requesting devices 100 and responding devices to communicate and report memory block status more efficiently. Typically, memory 102 or a memory block of a responsive device's memory is a collection of resources (eg, bytes, words, packet slots) that can be allocated among multiple concurrent operations. Beneficially, the negotiation process allows responding devices to efficiently report memory block status compared to memory registration or reservation. In operation, a negotiation process is initiated for a memory block size based on one or more of bandwidth criteria, responding device capabilities, and requesting device capabilities. Since data communication between two devices, the requesting device 100 and the responding device, depends on one or more parameters, such as bandwidth, i.e., the maximum rate of data transmission across a given path (e.g., communication interface 106); the responding device and the requesting device The capabilities of the device, such as processing power, size of memory, data transfer rate of the device, etc. Therefore, taking into account all these parameters, a negotiation process is initiated to effectively determine the favorable outcome for each party (or computing device). The term "favorable outcome" refers to the outcome of the consultation process. A beneficial result may be the size determined for the associated memory block.

In one embodiment, the controller 104 is further configured to receive an indication of an updated memory block size. Based on the negotiation process, the responding device uses the updated memory block size to transmit. Notably, the responding device is configured to change the memory block size used to report the memory block status based on requests from the requesting device 100 . Responding devices may decide to report different memory block sizes. The requesting device 100 or controller 104 may select any size or granularity of memory blocks based on requirements. Notably, the controller 104 dynamically changes the granularity (or block size) during data exchange to allow flexibility due to the number of pages being swapped out and to improve network utilization. Additionally, the responding device is enabled to allow opt-in, opt-out notifications for a specific requesting device, such as the requesting device 100, where the term "specific requesting device" refers to the requesting device 100 currently using a memory region or block of memory. Upon receiving the updated size, the requesting device 100 or controller 104 may determine whether the memory block to be accessed is ready.

In another embodiment, the updated memory block size is based on one or more of the number of memory actions, bandwidth usage, responding device capabilities, and requesting device 100 capabilities. The updated memory block size is based on the number of memory actions or operations such as RDMA reads, RDMA writes, etc. The number of memory actions performed to complete data communication is also considered when determining the updated memory block size. Additionally, other factors, such as bandwidth usage or bit rate, are used to determine the amount of data for a potential communication, responding device capabilities and requesting device 100 capabilities, ie, the physical configuration and capabilities of each of the responding device and requesting device 100. Typically, the capabilities of the requesting device 100 and the responding device are similar or equal.

In one embodiment, the controller 104 is also configured to store the memory status of the memory block. Based on the received notification or the negotiation process, the controller 104 is configured to store the memory state of one or more memory blocks in the memory 102 . In other words, the controller 104 keeps track of each of one or more memory blocks. Storing information regarding the memory status of one or more blocks in the responding device's memory enables the requesting device 100 to perform future operations. For example, if a memory block is determined to be ready for access by the controller 104, the requesting device 100 may request access to the memory block using the stored memory state of the memory block of the responding device.

In one embodiment, the controller 104 is also configured to group multiple memory blocks into memory block groups. Based on this implementation, the controller 104 may group multiple memory blocks into memory block groups. A "memory block group" refers to a collection of memory blocks. The grouping of memory blocks is intended to make data communication more efficient by grouping memory blocks used for similar operations or functions and processing them collectively. In addition, the controller 104 is configured to determine the memory group status of the memory block group based on the memory status of the plurality of memory blocks. Based on the information received from the notification and/or negotiation process, the controller collectively determines the memory group status of multiple memory blocks. Preferably, memory blocks of one type are grouped together by the controller 104. For example, memory blocks that are smaller than a desired threshold are grouped together, and memory blocks that are larger than a desired threshold are grouped together to determine the memory block group status. Additionally, the controller 104 is configured to store memory bank status in the memory 102 . The memory block status may be used to determine the controller's 104 accessibility to a memory block or memory block group.

In another embodiment, the memory group status includes a counter, and the controller 104 is configured to adjust the counter if the received memory status indicates that a memory block included in a plurality of memory blocks of the memory block group has changed its memory status. "Counter" refers to a register that keeps track of the memory status of multiple memory blocks. Typically, the counter is incremented or decremented by a value based on the memory state of each memory block. For example, if the memory status of the associated memory block is ready or accessible, the counter will increment its value. In another example, if the associated memory block is not ready or inaccessible, the counter will be decremented. It is worth noting that the counter determines the number of available and unavailable memory blocks in the memory block group.

In yet another embodiment, the memory group status includes a memory flag, and the controller 104 is configured to set the memory flag if the notification indicating the memory status indicates that the memory blocks of the memory block group are not currently in the memory of the responding device. A "memory flag" is a flag that has a value that is a function or process signal. Typically, a memory flag is a binary flag containing a Boolean value such as true or false. However, not all memory flags are binary and may store a range of values. It is worth noting that the value of the memory flag is used to determine the memory status of the memory block group. The memory flag indicates whether any memory block of the memory block group is currently in the memory of the responding device, that is, whether the memory block of the memory block is accessible or ready. Therefore, counters are used to determine the memory status of a group of memory blocks based on the values of the memory flags associated with the memory blocks individually and collectively.

In one embodiment, notifications about memory status indicate changes to one or more memory blocks. The responding device may operate with one or more parties or computing devices or perform one or more operations concurrently with the requesting device 100 . Thus, a memory block may be used by other operations or other devices, for example under priority, or a previously unavailable data block may become available. In this case, the responding device is configured to transmit a notification including a memory state indicating a change in the memory state of one or more memory blocks. For example, if a previously accessible memory block has been exhausted, the responding device transmits an updated unavailable memory status via a notification. In another example, if a previously unavailable memory block is now available for implementation, the responding device transmits the updated availability status to the requesting device 100 via a notification. Typically, changes in the state of any one of one or more memory blocks are indicated by notifications from the responding device. For example, a memory block can change from the swapped-out state to the swapped-in state, or vice versa.

In one embodiment, notifications about memory status indicate the current memory status of one or more memory blocks. In situations where the operation may be delayed or completed at a later time, the requesting device 100 may request the current state of the memory block. In this case, the controller 104 may request the responding device to obtain the current status related to one or more memory blocks. Therefore, the responding device is used to transmit notifications that include the current memory state of one or more blocks. Beneficially, the current state prevents failures that may occur during operation when a memory block is unavailable.

In another embodiment, the controller 104 is further configured to send a message including a prefetch hint request indicating the memory block to be accessed to the responding device. "Message" refers to another type of message, such as a notification to enable data communication between two computing devices, such as the requesting device 100 and the responding device. A "prefetch hint request" refers to the type of request transmitted before actual data communication to gather knowledge about the state of a memory region and/or memory block. A prefetch hint request indicates to the responding device the memory block to be accessed. Typically, if the memory region or memory block to be accessed indicated in the notification is not ready yet, a prefetch hint request indicating the memory block to be accessed by the requesting device 100 is transmitted. In other words, if it is determined that the memory block to be accessed is not yet ready (or swapped out) in the memory of the responding device, a message including a prefetch hint request indicating the memory block to be accessed (or swapped in) is transmitted to the responding device. The information received from the notification is used to determine the prefetch hint request. The prefetch hint request includes at least the size, location, structure, or address of the memory block and/or memory region to be accessed. Beneficially, prefetch hint requests save time lost waiting for a response compared to individual request operations. If the requested memory area or memory block is available, the controller 104 avoids transmitting the request or message. If the requested memory region or memory block is not available, the requesting device 100 may select another memory region where other memory regions and memory blocks are available or ready for access.

In one embodiment, the notification regarding memory status indicates that the responding device is incompatible. In general, where more memory blocks cannot be implemented, the responding device should be considered incompatible. Typically, the notification is at least one of a memory status error notification, a memory status unavailable notification, or a memory status reset notification. The term "incompatible" refers to the state of a responding device when the responding device is unable to satisfactorily accommodate requests from the requesting device 100 . In this case, the responding device transmits a notification to the requesting device 100, such as a memory status error or memory status unavailability. Here, the controller 104 is also configured to adjust requests related to one or more memory blocks to be accessed accordingly. The controller 104, upon receiving a notification from a responding device, is configured to adjust requests related to one or more memory blocks to be accessed to suit the conditions of the responding device indicated by the notification. Typically, if the memory status related to one or more memory blocks to be accessed indicates any error condition or the memory status cannot be transferred further, then in this case the requesting device is used to clear the database (or tracker). The database used to store the memory state of one or more memory blocks is cleared, and information accumulated before the database was cleared is not used for further decisions related to memory requests. Optionally, in this case where the requesting device 100 or the controller 104 is configured to adjust the memory request, the controller 104 is configured not to send a message including the prefetch hint request. Since the information about the memory blocks in the memory 102 is irrelevant, the controller 104 does not transmit the message including the prefetch hint request.

Upon receiving the notification, the controller 104 is configured to determine whether the memory block to be accessed is ready in the memory of the responding device based on the memory status. Controller 104 receives notifications through communication interface 106 and analyzes notifications received from responding devices. Based on the notification, the controller 104 obtains knowledge about the memory status of the memory area and/or memory block to be accessed, and determines whether the memory block to be accessed is ready. Based on one or more parameters such as the available size of the memory area or block, the location of the memory area or address, the controller 104 determines the accessibility of the memory block and further performs necessary operations. If the memory area requested in the request is available, the controller 104 may transmit another message, prompt, or request to the responding device.

After determining the accessibility of the memory block based on the notification or memory status, the controller 104 is configured to transmit the request to the responding device through the communication interface 106 . If it is detected through the memory status that a memory area or memory block is ready to be accessed, a request related to the memory block is transmitted by the controller 104 . The controller 104 transmits the request, including but not limited to the required memory block size, the location or address of the memory block, the location or address of the memory block to be accessed by the requesting device 100 . After analyzing or checking the memory of the responding device, a request is sent to the responding device to begin the actual data to be transferred.

Referring now to FIG. 2 , shown is a flowchart of a method 200 for requesting the device 100 to perform an RDMA operation according to an embodiment of the present invention. Figure 2 should be read in conjunction with Figure 1. For example, method 200 may be performed by controller 104 of requesting device 100, which controller is shown and explained in conjunction with FIG. 1 . As shown, method 200 for requesting a device, such as requesting device 100, includes steps 202-210.

At step 202, method 200 includes associating with a memory area of the responsive device, the memory area including one or more memory blocks. A memory area is a portion of memory to be associated with a responsive device.

At step 204 , the method 200 includes receiving a notification regarding the memory status from the responding device via the communication interface 106 . The controller 104 receives the notification from the responding device after analyzing the memory area and memory block to determine the memory status.

At step 206, the method 200 includes determining whether the memory block to be accessed is ready in the memory of the responding device based on the memory status. The controller 104 is configured to determine whether a memory block is ready to be accessed based on the memory status of each memory block.

At step 208, the method 200 includes transmitting a request to the responding device through the communication interface, the request related to the memory block. Controller 104 is configured to transmit requests to access memory blocks or memory regions based on memory status and notifications received from responding devices.

Steps 202 and 208 are illustrative only, and other alternatives may be provided in which one or more steps are added, one or more steps are deleted, or one or more steps are provided in a different order without departing from the claims herein. The scope of the book.

The present invention also provides a computer program product, including a non-transitory computer-readable storage medium having computer program code stored thereon, and the computer program code can be executed by the controller 104 to perform the method 200. Generally, method 200 is used for a requesting device (such as requesting device 100 shown and explained in connection with Figure 1) to communicate with a responding device to perform an RDMA operation. Examples of implementations of non-transitory computer-readable memory modules include, but are not limited to, electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), Read only memory (ROM), hard disk drive (HDD), flash memory, secure digital (SD) card, solid-state drive (SSD), computer-readable storage media and/or CPU cache memory. Computer-readable storage media used to provide non-transitory memory may include, but are not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing.

Referring now to Figure 3, shown is a block diagram of a response device 300 in accordance with an embodiment of the present invention. The responding device 300 should be read in conjunction with the requesting device 100 of FIG. 1 . Typically, the responding device 300 communicates with the requesting device 100 to perform RDMA operations based on the transmitted messages and notifications.

As shown, response device 300 includes memory 302, controller 304, and communication interface 306. The memory 302 and communication interface 306 are similar to the memory 102 and communication interface 106 of the requesting device 100 of FIG. 1 and therefore explanation thereof is avoided for the sake of brevity. Controller 304 includes suitable logic, circuitry, and/or interfaces required to perform RDMA operations. Controller 304 is a computing element for processing instructions that drive responsive device 300 . Examples of the controller 304 include, but are not limited to, a network interface controller, a microprocessor, a microcontroller, a complex instruction set computing (CISC) microprocessor, a reduced instruction set (RISC) microprocessor Or a very long instruction word (VLIW) microprocessor. Alternatively, one or more controllers, processing devices and elements similar to controller 304 may be arranged in various architectures for responding to and processing instructions driving responsive device 300 . The responding device 300 manages memory by pages/frames (controlled by the OS) and the agreed memory block size is used to control the amount of notifications sent over the network so that the size does not need to be stored in the responder or responding 300's database.

The controller 304 is used to associate a memory area with the requesting device 100 . Controller 304 associates the requested memory region with requesting device 100 based on the request from requesting device 100 . A "memory region" refers to a region, also called a resource partition, extent, or memory context, which is a collection of allocated blocks that can be efficiently freed based on implementation. As used herein, a memory area is a portion of a larger memory such as memory 302. Typically, all blocks in a memory region are allocated in a contiguous memory address range or ranges of memory addresses, similar to traditional stack frame allocation. Memory 302 includes one or more memory areas that may be allocated to different operations based on implementation. The request includes at least the size of the memory area required or requested by the requesting device 100 . The controller 304 is used to associate memory areas. In one implementation, the memory area will be associated by the controller 304 with the memory 302 of the responsive device 300 . The memory 302 of the responding device 300 is similar to the memory 102 of the requesting device 100 . Typically, a memory region associated with either memory 102 or 302 includes one or more memory blocks. Typically, the responding device 300 associates the memory region using information obtained from requests related to the memory region, such as from size, address, type, etc. It should be understood that association and memory registration can be implemented in other different ways without limiting the scope of the invention.

Controller 304 is used to determine memory status related to the memory area. "Memory State" means the state of an associated memory region, which may or may not be available or ready at the time of request, including one or more memory blocks. After associating the memory area, the controller 304 analyzes the memory 302 to determine the memory status associated with the memory area. For example, the controller 304 checks the memory area based on the size of the memory area indicated in the request from the requesting device 100 . So, basically, each of one or more memory blocks is screened or analyzed to individually determine their status. The collective state of each of the one or more memory blocks that form a memory region is called the memory state. Typically, memory status indicates whether the requested memory or memory region is available.

The controller 304 is configured to transmit notifications regarding memory status to the requesting device 100 through the communication interface 306 . "Notification" refers to the type of message sent from the responding device 300. Optionally, the notification can be transmitted after a preconfigured period of time. Additionally, optionally, the notification may be transmitted when the accumulation of notification events exceeds a predetermined threshold. "Scheduled Threshold" refers to the maximum number of notification events that are queued or accumulated before a notification is transmitted. The controller 304 transmits the notification after determining the memory status associated with the memory area. The memory status indicates whether the requested memory area and the memory blocks therein are available for use, and the controller 304 then transmits a notification about the memory status to the requesting device 100 . Typically, the notification relates to a memory region, however, the notification may include the memory status of each of the associated one or more memory blocks in one or more memory regions in memory 302 . Based on the memory status, the notification indicates a positive acknowledgment regarding the memory region indicating the memory region's availability for operation, or a negative acknowledgment indicating the memory region's unavailability.

In another embodiment, the controller 304 is also configured to transmit an indication of the memory block size, where the memory block size is less than or equal to the memory area. A memory block is a portion of the memory area of the response device 300 . Controller 304 transmits an indication of the memory block size of responding device 300 to requesting device 100 . Depending on the implementation, the indication may include the size of one or more memory blocks. Based on the transmitted indication, the requesting device 100 or the controller 104 may reconfigure or delay the RDMA operation by delaying the transmission of some data packets being transmitted based on the memory status of the corresponding memory block. For example, the controller 304 transmits an indication that the memory block size is 30KB, based on which the requesting device 100 may reconfigure or set the granularity of notifications related to one or more different memory blocks. It is worth noting that the memory block size is smaller than the size of the memory region. In general, any requesting device, such as requesting device 100, may require memory blocks or memory regions of different sizes for different implementations and efficient data communication. Different sizes of memory blocks or memory regions provide granularity to memory 102 and thus flexibility to the memory. Therefore, to avoid any discrepancies that may inhibit efficient data communication during operation, the responding device 300 transmits the status of the memory block to assist the requesting device 100 in determining the accessibility of the associated memory block. Additionally, notifications about memory status relate to one or more memory blocks. Typically, each of one or more memory blocks includes a separate memory state that indicates the status or availability of the memory block. Overall, each memory block helps determine the memory state of the memory region.

In another embodiment, the controller 304 is further configured to receive the indicated memory block size from the requesting device prior to transmitting the indication of the memory block size. "Indicated size" refers to an indication of the granularity of the memory status notification provided to the responding device 300. The controller 304 receives the indicated memory block size, where the indicated size is based on the memory block size in the memory 102 of the requesting device 100 .

In yet another embodiment, the controller 304 is further configured to initiate a negotiation process with the requesting device 100 for a memory block size based on one or more of bandwidth criteria, responding device 300 capabilities, and requesting device 100 capabilities. . The controller 304 is used to initiate a negotiation process with the controller 104 or the requesting device 100. "Negotiation process" refers to a simple protocol that provides a way to mutually decide and agree on a value for a memory block size. Alternatively, the controller 104 of the requesting device 100 is also used to initiate a negotiation process with the controller 304 of the responding device 300 . The bidirectional nature of the negotiation process allows requesting device 100 and responding device 300 to efficiently communicate memory block sizes. Typically, a memory block of memory 102 of requesting device 100 or memory 302 of responding device 300 is a collection of resources (eg, bytes, words, packet slots) that can be allocated among multiple concurrent operations. Advantageously, the negotiation process allows the responding device 300 to efficiently report the status of the memory or memory block. In operation, a negotiation process is initiated for a memory block size based on one or more of bandwidth criteria, responding device 300 capabilities, and requesting device 100 capabilities. In other words, negotiation is established for large and small granularities. Since data communication between two devices, the requesting device 100 and the responding device 300, depends on one or more parameters, such as bandwidth, i.e., the maximum rate of transmission across a given path (e.g., the communication interface 106); the responding device 300 and the requesting device 300 The capabilities of the device 100, such as processing power, size of memory, data transfer rate of the device, etc. Therefore, taking into account all these parameters, a negotiation process is initiated to effectively determine the favorable outcome for each party (or computing device). The term "favorable outcome" refers to the outcome of the consultation process. An advantageous result may be the determined size of the memory block for which the memory state is built.

In one embodiment, the controller 304 is also configured to receive an indication of an updated memory block size. According to the negotiation process, the responding device 300 is configured to receive the updated memory block size. Notably, the responding device 300 is used to change the memory block size based on the requesting device 100 . The response device 300 or the controller 304 may select any size or granularity of memory blocks based on requirements. Notably, the controller 304 dynamically changes the granularity (or block size) during data exchange to allow flexibility due to the number of pages being swapped out and to improve network utilization. Additionally, the responding device 300 is enabled to allow opt-in, opt-out notifications for specific requesting devices, such as the requesting device 100, where the term "specific requesting device" refers to the requesting device 100 that actually uses the memory area or block of memory. Upon receiving the updated status, the requesting device 100 or controller 104 may determine whether the memory block to be accessed is ready.

In another embodiment, the updated memory block size is based on one or more of the number of memory actions, bandwidth usage, responding device 300 capabilities, and requesting device 100 capabilities. The updated memory block size is based on the number of memory actions or operations such as RDMA reads, RDMA writes, etc. The number of memory actions performed to complete data communication is also considered when determining the updated memory block size. Additionally, other factors, such as bandwidth usage or bit rate, are used to determine the amount of data for a potential communication, responding device 300 capabilities and requesting device 100 capabilities, ie, the physical configuration and capabilities of each of the responding device 300 and requesting device 100 . Typically, the capabilities of the requesting device 100 and the responding device 300 are similar or equal.

In one embodiment, the controller 304 is also configured to determine the memory status of one of the one or more memory blocks. Typically, controller 304 is used to determine memory status associated with a memory region. However, the controller 304 is used to determine the memory status of one of one or more memory blocks in the memory area. Specifically, the controller 304 determines the memory status of the memory blocks required for data communication in the associated memory area. Controller 304 may be used to determine the memory status of one, two, more than two, or all of one or more memory blocks. When determining the memory state of a memory region, the collective memory state of one or more memory blocks can be considered.

In another embodiment, the controller 304 is also configured to determine the memory status of a plurality of one or more memory blocks. The controller 304 is configured to determine the memory status of a plurality of one or more memory blocks in the memory area. Specifically, a memory area including one or more memory blocks is grouped into multiple memory blocks. Controller 304 is used to determine the status of each of the plurality of memory blocks. In determining memory status based on requests from requesting device 100 or responding device 300, the determined status of multiple memory blocks may be selected individually or as a group. Typically, multiple memory blocks associated with the requesting device 100 allow the size of the database used to store notifications to be reduced.

In one embodiment, the controller 304 is further configured to determine the memory status of the memory block based on changes in the memory status of the memory block. The memory state of a memory block is determined based on changes in the memory state of the memory block. Here, the change is attributed to mapped and unmapped memory blocks and corresponding changes in the memory state of the memory blocks. Responding device 300 may operate with one or more parties or computing devices to perform one or more operations concurrently with requesting device 100 . Accordingly, one or more memory blocks in the response device 300 may be used or occupied by other operations or other devices, for example, under priority, or data blocks that were previously unavailable may become available. In this case, the controller 304 is configured to transmit a notification including a memory status indicating a change in the memory status of one or more memory blocks. For example, if a previously accessible memory block has been exhausted, the controller 304 transmits an updated unavailable memory status via a notification. In another example, if a previously unavailable memory block is now accessible, the controller 304 transmits the updated available memory status to the requesting device 100 via a notification. Typically, changes in the state of any one of one or more memory blocks are indicated by notifications from the responding device 300 . Typically, the responding device 300 is used to notify the requesting device 100 of the selected granularity. In one example, when the number of unmapped pages or memory blocks of a memory region increases from zero to one or more, a notification is sent to the requesting device 100 indicating that the block may be unmapped. In another example, when the number of unmapped pages for a block decreases from one or more to zero, a notification is sent to the requesting device that the block may be mapped.

In one embodiment, the controller 304 is further configured to determine the memory status of the memory block based on determining the memory location of the memory block. Controller 304 determines the location of the memory block before determining the memory status. "Memory location" refers to an address such as a virtual address (VA) or a physical address (PA) of a memory block. The location of the memory block may be provided by the requesting device 100 in a prefetch hint request. Once the location of the memory block is determined by controller 304 , controller 304 associates the memory block with the determined memory location in memory 302 . Such correlation enables the responding device 300 or controller 304 to efficiently determine memory status based on the memory location. Optionally, the controller 304 analyzes the requested memory block and determines the memory status of the memory block after associating the memory status based on the memory location of the memory block. Specifically, whether the memory location is available or swapped in or out. Furthermore, such a protocol allows active determination of memory status on request.

In one embodiment, the controller 304 is further configured to determine the memory status of one or more memory blocks within a time period, and when the time period expires, transmit a notification indicating the memory status of the one or more memory blocks. . The response device 300 or controller 304 is used to determine the memory status of one or more blocks based on a threshold for swapping in or out of memory blocks, a time period, or a maximum notification reported per message. Notifications may be scheduled to be implemented or transmitted over a regular period of time based on demand or as a preferred number of memory swap operations in a given time period, or both. Therefore, when the memory status is updated or changed, the notification is scheduled to be sent at a preset time after the notification change. The preset time is, for example, 1 millisecond (millisecond, ms), 10ms, 100ms, 1 second (second, s), 2s , 10s, etc. Additionally, if the time period expires, the controller 304 is configured to transmit a notification indicating the memory status of the one or more memory blocks. In one example, if a notification is already scheduled, append the memory state change to the exit notification.

In one embodiment, the controller 304 is also configured to determine the memory status of one or more memory blocks in the memory 302 and determine the number of memory actions. Additionally, based on the determined number of memory actions, the controller 304 is configured to determine whether the number of memory actions exceeds a threshold number, where the term "threshold number" refers to a predetermined value of the number of memory actions. The controller 304 is also configured to transmit notifications indicating the memory status of one or more memory blocks. In one example, if the number of memory actions exceeds a threshold number of memory actions, the controller 304 transmits the notification immediately, ie, before the time period. In another example, if the number of memory actions does not exceed a threshold number, controller 304 waits for a period of time before transmitting the notification.

In one embodiment, the controller 304 is further configured to receive a prefetch hint request from the requesting device 100 through the communication interface 306, the prefetch hint request indicating one or more memory blocks. A "prefetch hint request" refers to the type of request transmitted before actual data communication to gather knowledge about the state of a memory region and/or memory block. The prefetch hint request indicates a block of memory to be loaded into/mapped into the memory 302 of the responding device 300 . Typically, if the memory region or memory block to be accessed on request during association is not ready, a prefetch hint request is transmitted indicating the memory block to be accessed by the requesting device 100. In other words, if it is determined that the memory block to be accessed is not ready (or swapped out) in the memory 302 of the responding device 300 and the indicated memory block or blocks are loaded into the memory 302, then a transmission is made to the responding device 300 Contains a message indicating a prefetch hint request for a memory block to be accessed (or swapped in). The information received from the notification is used to determine the prefetch hint request. The prefetch hint request includes at least one of the size, structure, location, or address of the memory block and/or memory region to be accessed. Beneficially, this message saves time lost waiting for a response compared to a separate request operation. In one example, the controller 304 may transmit a notification regarding the memory status if the requested memory region or memory block is available or swapped in.

Referring now to FIG. 4 , a flowchart of a method 400 for performing an RDMA operation in response to a device 300 is shown according to an embodiment of the present invention. Figure 4 should be read in conjunction with Figure 3. For example, method 400 may be performed by controller 304 of response device 300, which controller is shown and explained in connection with FIG. 3 . As shown, method 400 for a responding device, such as responding device 300, includes steps 402-406.

At step 402 , method 400 includes associating a memory region with requesting device 100 . Controller 304 associates the memory area requested from requesting device 100 with a memory area in memory 302 .

At step 404, method 400 includes determining memory status associated with the memory region. Controller 304 is configured to determine memory status associated with a memory region in one or more ways, where the memory status indicates whether a block of memory is ready to be accessed (or swapped in) or not yet ready to be accessed (swapped out).

At step 406 , the method 400 includes transmitting a notification regarding the memory status to the requesting device 100 via the communication interface 306 . The controller 304 is configured to transmit the notification, where the notification at least includes the memory status of the memory area or memory block.

Steps 402 through 406 are illustrative only, and other alternatives may be provided in which one or more steps are added, one or more steps are deleted, or one or more steps are provided in a different order without departing from the claims herein. The scope of the book.

The present invention also provides a computer program product, including a non-transitory computer-readable storage medium having computer program code stored thereon, and the computer program code can be executed by the controller 304 to perform the method 400. Generally, method 400 is used for a responding device (eg, responding device 300 shown and explained in connection with FIG. 3) to communicate with a requesting device (eg, requesting device 100 shown and explained in connection with FIG. 3) for performing execution based on a memory state. RDMA operations. Examples of implementations of non-transitory computer-readable memory modules include, but are not limited to, electrically erasable programmable read-only memory (EEPROM), random access memory (random access memory, RAM), read-only memory (read only memory, ROM), hard disk drive (HDD), flash memory, secure digital (SD) card, solid-state drive (SSD), computer-readable storage media and/or CPU cache memory. Computer-readable storage media used to provide non-transitory memory may include, but are not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing.

Referring to Figure 5, there is shown an RDMA system 500 depicting performing RDMA operations in accordance with various embodiments of the present invention. Figure 5, which should also be read in conjunction with Figures 1-4, depicts a requesting device 100 and a responding device 300. As shown, at step 502 , the requesting device 100 includes the controller 104 for transmitting a request 510 to associate a memory region of the responding device 300 through the communication interface 106 (not shown). At step 504, the responding device 300 associates the memory region with the memory 302 (not shown) of the responding device 300 and the memory region including one or more memory blocks. Additionally, the response device 300, through the controller 304, is used to determine memory status associated with the memory region. After determining the memory status, the controller 304 is configured to transmit a notification 512 in response to the request 510 from the requesting device 100 . Additionally, the controller 104 receives notifications 512 from the responding device 300 regarding memory status related to the memory region via the communication interface 306 (not shown). In addition, at step 506, the controller 104 is configured to determine whether the memory block to be accessed is ready in the memory 302 (not shown) of the responding device 300 based on the memory status, and transmit a request 514 to the responding device 300 through the communication interface 106, requesting 514 is related to memory blocks.

A schematic flowchart depicting an exemplary scenario of an embodiment of an RDMA system 600, 700 is shown with reference to Figures 6 and 7, in accordance with various embodiments of the present invention. The RDMA systems 600, 700 of Figures 6-7 should also be read in conjunction with Figures 1-4, depicting the requesting device 100 and the responding device 300. As shown in Figure 6, RDMA system 600 includes a requesting device, such as requesting device 100, which is shown and explained in conjunction with Figures 1-2. RDMA system 600 also includes a response device, such as response device 300, which is shown and explained in conjunction with Figures 3-4. Similarly, as shown in Figure 7, RDMA system 700 includes a single requesting device, such as requesting device 100, and a single responding device, such as requesting device 300.

In one embodiment, as shown in Figure 6, in operation, an RDMA system 600 including a requesting device 100 communicates with a responding device 300 in which memory regions or memory blocks are mapped or available. At step 602, the responding device 300 checks the memory status of the memory area and transmits a notification 610 related to the memory area. Typically, the responding device 300 detects whether the memory region is partially mapped, fully mapped, or unmapped, and transmits a notification 610 to the requesting device 100 accordingly. In operation, the response device 300 transmits a notification 610 of an event regarding a page frame number (PFN) associated with a memory region. The requesting device 100, upon receiving the notification 610, marks the memory region as "potentially mapped". After analyzing the notification for a period of time, the requesting device 100 may perform one or more actions. As shown, at step 604, the requesting device 100 issues an RDMA read request 612 for the memory area. When a memory region is already mapped, the requesting device 100 readily detects that the memory region is "potentially mapped" and therefore transmits the request immediately without transmitting a prefetch hint request and without waiting for a prefetch period before transmitting the request. At final step 606, the responding device 300 transmits an RDMA read response 614 to the requesting device 100.

In one embodiment, as shown in Figure 7, in operation, the RDMA system 700 including the requesting device 100 communicates with the responding device 300 in which a memory region or memory block is unmapped or unavailable. At step 702, the responding device 300 checks the memory status of the memory area and transmits a notification 712 related to the memory area if the memory area is unmapped. Typically, the responding device 300 detects and marks the memory region as potentially unmapped, and transmits a notification 712 to the requesting device 100 accordingly. In operation, the response device 300 transmits a notification 712 of an event regarding a page frame number (PFN) associated with a memory region. At step 704, the requesting device 100, upon receiving the notification 712, marks the memory region as "potentially unmapped." After analyzing the notification 712 for a period of time, the requesting device 100 may perform one or more actions. At step 704, the requesting device 100 detects that the memory region is "potentially unmapped" and therefore transmits a prefetch hint request 714 and waits for a prefetch period until the memory region is fully mapped. At step 706, the responding device performs a swap operation on the unavailable memory block to completely map the memory area. After waiting for the prefetch period, at step 708, the requesting device 100 issues an RDMA read request 716 (eg, request 612) for the memory region. Since the memory area is now fully mapped, in final step 710 the responding device 300 transmits an RDMA read response 718 to the requesting device 100.

Allocate memory based on size, or request an operation based on size change. Typically, the responding device 300 determines the current state of the shared virtual memory area or shared virtual address before receiving the data packet. If, SVA or SVM is supported and enabled. Then, the incoming memory data packet is transmitted to the input-output memory management unit (IOMMU) or system memory management unit (SMMU) as a DMA transaction, and then converted according to method 400. Otherwise, if SVA or SVM is not supported or enabled, the responding device 300 or NIC (Network Interface Controller) checks the MTT (Memory Translation Table) and performs translation according to method 400 to determine the memory address to be accessed and the memory address to be set. Page resolution or granularity. The term "MTT or memory translation table" refers to the data structure used by any virtual memory system to store mappings between virtual addresses and physical addresses. Specifically, virtual addresses are used by programs executed on a computing device such as response device 300, while physical addresses are used by hardware, or more specifically, by the RAM subsystem. MTT is a key component for memory address translation, which is necessary to access data in memory 302. Here, the memory block enables the system to set the resolution to be tracked. Typically, the resolution or granularity depends on available memory (eg, memory 302), the exchange rate of memory blocks or memory regions on the response device 300, the service level agreement (SLA) of the application, and so on.

Modifications may be made to the embodiments of the invention described above without departing from the scope of the invention as defined in the appended claims. Expressions such as "comprises," "includes," "combines," "has," and "is" used to describe and claim the present invention are to be construed in a non-exclusive manner, allowing for items, portions or elements not expressly described can also appear. References to the singular shall also be construed as relating to the plural. As used herein, the word "exemplary" means "serving as an example, instance, or illustration." Any embodiment described as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments and/or does not exclude combination of features of other embodiments. As used herein, the word "optionally" means "provided in some embodiments and not provided in other embodiments." It is to be understood that certain features of the invention, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, for the sake of brevity, various features of the invention that are described in the context of a single embodiment may also be provided separately or in any suitable combination, or as may be suitably provided in any other described embodiment of the invention.

Claims (31)

1. A requesting device (100), characterized by comprising a memory (102), a controller (104) and a communication interface (106), the controller (104) being used for: Associating with a memory area of the responding device (300), the memory area including one or more memory blocks; receiving notification (512) from the responding device via the communication interface (106) regarding memory status related to the memory region; Determine whether the memory block to be accessed is ready in the memory (302) of the responding device (300) based on the memory status; A request (514) is transmitted to the responding device via the communication interface, the request (514) relating to the memory block to be accessed. 2. The requesting device (100) according to claim 1, characterized in that the controller (104) is further configured to receive an indication of a memory block size, the memory block size being less than or equal to the size of the memory area. , wherein the notification (512) about the memory status relates to one or more memory blocks. 3. The requesting device (100) according to claim 2, characterized in that the controller (100) is further configured to send a request to the responding device (300) before receiving the indication of the memory block size. Transfer the indicated memory block size. 4. The requesting device (100) according to claim 2, characterized in that the controller (104) is further configured to initiate a negotiation process with the responding device (300) for the memory block size, wherein, The memory block size is based on one or more of bandwidth criteria, responding device (300) capabilities, and requesting device (100) capabilities. 5. The requesting device (100) according to any one of claims 2 to 4, characterized in that the controller (104) is further configured to receive an indication of an updated memory block size. 6. The requesting device (100) according to claim 5, characterized in that the updated memory block size is based on one of the number of memory actions, bandwidth usage, responding device (300) capabilities and requesting device (100) capabilities or more. 7. The requesting device (100) according to any one of the preceding claims, characterized in that the controller (104) is further configured to store the memory state of a memory block. 8. The requesting device (100) according to any one of the preceding claims, characterized in that the controller (104) is further configured to: Group multiple memory blocks into memory block groups; determining a memory group status of the memory block group based on the memory status of the plurality of memory blocks; Stores the memory bank state. 9. The requesting device (100) of claim 8, wherein the memory group status includes a counter, and the controller (104) is configured to: if the received memory status indicates that the memory block If a memory block included in the plurality of memory blocks of the group has changed its memory state, the counter is adjusted. 10. The requesting device (100) of claim 9, wherein the memory bank status includes a memory flag, and the controller (104) is configured to: if the notification (512) indicating the memory status indicates The flag is set if a memory block of the memory block group is not currently in the memory (302) of the responding device (300). 11. The requesting device (100) according to any one of the preceding claims, characterized in that said notification (512) about memory status indicates a change of one or more memory blocks. 12. The requesting device (100) according to any one of the preceding claims, characterized in that said notification (512) about memory status indicates the current memory status of one or more memory blocks. 13. The requesting device (100) according to claim 1, characterized in that the controller (104) is further configured to: if it is determined that the memory block to be accessed is in the memory (300) of the responding device (300) 302) is not ready, send a message including a prefetch hint request (714), which indicates the memory block to be accessed to the responding device (300). 14. The requesting device (100) according to any one of the preceding claims, characterized in that the notification (512) regarding memory status indicates that the responding device (300) is incompatible, wherein the controller ( 104)Also used in: Adjusting said request in relation to said memory block to be accessed accordingly, when appended to claim 13, The message including the prefetch hint request (714) is not sent. 15. A method (200) for requesting a device (100), characterized in that the requesting device (100) includes a memory (102), a controller (104) and a communication interface (106), the method ( 200) includes: Associating with a memory area of the responding device (300), the memory area including one or more memory blocks; receiving notifications (512) from the responding device (300) via the communication interface (106) regarding memory status; Determine whether the memory block to be accessed is ready in the memory of the responding device (302) based on the memory status; A request (514) is transmitted to the responding device via the communication interface, the request (514) relating to the memory block to be accessed. 16. A computer-readable medium carrying computer instructions, characterized in that the computer instructions, when loaded into and executed by a controller (104) of a requesting device (100), cause the requesting device (100) to The method (200) according to claim 15 can be implemented. 17. A response device (300), characterized by comprising a memory (302), a controller (304) and a communication interface (306), the controller (304) being used for: Associating a memory area with the requesting device (100), the memory area including one or more memory blocks; determining memory status associated with said memory region; A notification regarding the memory status is transmitted (512) to the requesting device (100) through the communication interface (306). 18. The response device (300) according to claim 17, characterized in that the controller (304) is further configured to transmit an indication of a memory block size, the memory block size being less than or equal to the size of the memory area. , wherein the notification (512) about the memory status relates to one or more memory blocks. 19. The responding device (300) of claim 18, wherein the controller (304) is further configured to receive a request from the requesting device (100) before transmitting the indication of the memory block size. Receives the indicated memory block size. 20. The responding device (300) according to claim 18, wherein the controller (304) is further configured to initiate a negotiation process with the requesting device (100) for the memory block size, wherein, The memory block size is based on one or more of bandwidth criteria, responding device (300) capabilities, and requesting device (100) capabilities. 21. The response device (300) according to any one of claims 18 to 20, wherein the controller (304) is further configured to receive an indication of an updated memory block size. 22. The responding device (300) of claim 21, wherein the updated memory block size is based on one of the number of memory actions, bandwidth usage, responding device (300) capabilities, and requesting device (100) capabilities. or more. 23. The response device (300) according to any one of claims 17 to 22, wherein the controller (304) is further configured to determine the value of one of the one or more memory blocks. memory status. 24. The response device (300) according to any one of claims 17 to 23, characterized in that the controller (304) is further configured to determine all the values of a plurality of the one or more memory blocks. Describes the memory status. 25. The response device (300) according to any one of claims 17 to 24, characterized in that the controller (304) is further configured to determine all the parameters of the memory block based on changes in the memory status of the memory block. Describes the memory status. 26. The response device (300) according to any one of claims 17 to 25, characterized in that the controller (304) is further configured to determine all the memory locations of the memory block based on determining the memory location of the memory block. Describes the memory status. 27. The response device (300) according to any one of claims 17 to 26, characterized in that the controller (304) is further configured to determine the response of one or more memory blocks within a time period. Memory status, upon expiration of the time period, transmit the notification indicating the memory status of the one or more memory blocks (512). 28. The response device (300) according to any one of claims 17 to 27, characterized in that the controller (304) is also used for: determining said memory status of one or more memory blocks; Determine the number of memory actions; Determine that the number of memory actions exceeds a threshold number; The notification indicating the memory status of the one or more memory blocks is transmitted (512). 29. The response device (300) according to any one of claims 17 to 28, characterized in that the controller (304) is also used for: A prefetch hint request (714) is received from the requesting device (100) through the communication interface (306), the prefetch hint request (714) indicates one or more memory blocks, and the indicated one or more memory blocks are A memory block is loaded into the memory (302). 30. A method (400) for a response device (300), characterized in that the response device (300) includes a memory (302), a controller (304) and a communication interface (306), and the method (300) 400) includes: Associate the memory area with the requesting device (100); determining memory status associated with said memory region; A notification regarding the memory status is transmitted (512) to the requesting device (100) through the communication interface (306). 31. A computer-readable medium carrying computer instructions, characterized in that the computer instructions, when loaded into and executed by a controller (304) of a response device (300), cause the response device (300) to The method (400) according to claim 30 can be implemented.