CN114675972B - Cloud network resource flexible scheduling method and system based on integral algorithm - Google Patents

Abstract

The invention discloses a cloud network resource flexible scheduling system and a cloud network resource flexible scheduling method based on an integral algorithm, wherein the cloud network resource flexible scheduling system and the cloud network resource flexible scheduling method comprise a virtual switch module, a data acquisition module, an integral calculation module and a resource scheduling module; the virtual switch module is used as a core module for bearing the network function of the VM and is used for providing network service for the VM according to the CPU cycle duty ratio; the data acquisition module is used for collecting the actual CPU clock cycle number and cycle duty ratio of the virtual switch consumed by each VM in each working time slice from the virtual switch module, and transmitting the actual CPU clock cycle number and cycle duty ratio to the integral calculation module; the integral calculation module is used for updating the integral value of each VM according to the input actual CPU clock cycle number and cycle duty ratio and transmitting the updated integral value to the resource scheduling module; the resource scheduling module is used for dynamically limiting the number of virtual switch CPU cycles and the cycle duty ratio consumed by each VM in the next working time slice according to the input integral value. The method ensures that a plurality of VM networks tend to reasonably level resource utilization.

Description

Cloud network resource elastic scheduling method and system based on integral algorithm

Technical Field

The present invention belongs to the technical field of cloud data center network, and in particular relates to a method and system for elastically scheduling virtualized cloud network resources based on an integral algorithm.

Background technique

With the development of information technology, the trend of informatization of industrial enterprises has become increasingly obvious. The development of many industrial enterprises has gradually tended towards intelligent operation, that is, enterprises going to the cloud. Enterprise going to the cloud refers to the whole process of enterprises using the Internet technology and cloud computing technology to connect social resources, shared platforms and work content, so as to carry out the construction of information management infrastructure, management methods, business processes, etc. As more and more enterprises are moving their businesses to the cloud, cloud service providers are bearing increasing operational pressure. On the one hand, the total network traffic of cloud service providers is gradually increasing; on the other hand, the service quality levels sold by cloud service providers are complex. In order to ensure that enterprise users can obtain a good service experience, cloud service providers must reasonably dispatch network resources according to traffic distribution characteristics and user preset service quality. Otherwise, the normal performance requirements of users will not be met, and it may also cause huge operational accidents such as local network hotspots causing cloud server CPU overload and downtime.

At present, mainstream cloud service providers all use virtual large Layer 2 network technology. Cloud service providers use virtualization technology to virtualize physical servers (hosts) into multiple virtual logical servers (VMs). For the convenience of business management and deployment, cloud service providers need to include a large number of VMs in the same Layer 2 broadcast domain. Since traditional VLAN Layer 2 technology cannot support tens of thousands or even hundreds of thousands of hosts in cloud data centers, cloud service providers have developed network protocols such as VXLAN, NVGRE, and STT to meet the requirements of large Layer 2 across regions and centers. The core component for implementing these protocols is the virtual switch. A virtual switch is a software application that allows network communication between VMs. It is usually deployed in the host system and works in conjunction with the virtual machine monitor (Hypervisor). Since the virtual switch is connected to the virtual network cards of multiple VMs through the software abstraction layer, it is a direct interactive component for VMs to communicate with the outside world on the network. It has a significant impact on the overall network performance of the VM cluster. Therefore, the elastic resource guarantee mechanism of the virtual switch is one of the important technologies for achieving service quality and performance isolation.

The elastic resource guarantee mechanism is to dynamically and reasonably schedule the computing resources consumed by the virtual switch in processing each VM network message. At present, virtual switches face the following challenges: First, because VMs deployed in the same host often have different service specifications, the virtual switch needs to limit the speed of each VM in multiple dimensions such as bit rate (BPS) and packet rate (PPS); second, the virtual switch needs to ensure isolation, that is, when one of the VMs generates abnormal traffic, the network service quality of other VMs on the same host is not affected; finally, the virtual switch needs to improve the overall utilization of the host computing resources as much as possible to avoid idle resources from being effectively utilized.

There are many algorithms and technologies for elastic resource guarantee in the industry. For example, FairCloud and NetShare based on the Fair sharing model; ElasticSwitch and Silo based on the Hose model, etc. In addition, Google has also proposed a PicNIC solution that allocates virtual switch CPU cycles based on the SLA corresponding to the VM. Although the above solutions have improved resource contention and service quality to varying degrees, these methods currently still lack fine-grained scheduling methods. A core issue is how to dynamically and seamlessly switch between two different working conditions: low load during idle time and high load during busy time.

Summary of the invention

In view of the above, an object of the present invention is to provide a cloud network resource elastic scheduling method and system based on an integral algorithm to ensure that resource utilization of multiple VM networks tends to a reasonable level.

To achieve the above-mentioned purpose of the invention, an embodiment provides a cloud network resource elastic scheduling system based on an integral algorithm, comprising a virtual switch module, a data acquisition module, an integral calculation module and a resource scheduling module;

The virtual switch module is used as a core module that carries VM network functions and is used to provide network services for VMs according to the CPU cycle ratio;

The data collection module is used to collect the actual number of CPU clock cycles and cycle proportion of the virtual switch consumed by each VM in each working time slice from the virtual switch module, and transmit it to the integral calculation module;

The integral calculation module is used to update the integral value of each VM according to the input actual CPU clock cycle number and cycle proportion, and transmit the updated integral value to the resource scheduling module;

The resource scheduling module is used to dynamically limit the number of virtual switch CPU cycles and cycle proportion consumed by each VM in the next working time slice according to the input integral value, so as to realize resource allocation.

In one embodiment, the virtual switch module includes a virtual network card interface module and a Netframe forwarding module;

The virtual network card interface module is used to provide an interface for network data communication with the VM;

The Netframe forwarding module is a user-mode network protocol stack component based on DPDK. As a core module for data forwarding, it is used to implement two-layer MAC address forwarding, that is, it is provided to the virtual network card interface module in the form of Netfilter Hook.

In one embodiment, the data collection module collects the number of virtual switch CPU clock cycles consumed by each VM in each working time slice and the proportion of the total number of cycles from the Netframe forwarding module.

In the integral calculation module of one embodiment, the process of updating the integral value of each VM includes:

Preset integral parameters BASE, MAX and MIN for each VM, where BASE is the basic consumption CPU cycle ratio, MAX and MIN are the maximum and minimum consumption CPU cycle ratios respectively;

For each VM, compare whether the consumed CPU cycle ratio is less than the BASE value. When the CPU cycle ratio is less than the BASE value, the points are increased by X1; when the CPU cycle ratio is greater than the BASE value, the points are reduced by X2 to obtain the points update result, where X1 and X2 are the preset increase ratio and decrease ratio. If X1 is less than X2, it means that the accumulation speed of points is slower than the consumption speed of points.

In the integral calculation module of one embodiment, when the integral value is 0, the integral is not updated and remains 0.

In the resource scheduling module of one embodiment, if the input integral value is 0, the maximum CPU cycle ratio consumed by the VM in the next second is set to BASE; if the integral value is greater than 0, the maximum CPU cycle ratio consumed by the VM in the next second is set to MAX; at the same time, it must be ensured that the minimum CPU cycle ratio consumed by the VM in the next second is MIN.

In one embodiment, the resource scheduling module dynamically controls the proportion of virtual switch CPU cycles consumed by each VM according to the input integral value and transmits it to the virtual switch module;

The virtual switch module provides network services of corresponding specifications for the VM according to the received CPU cycle ratio.

In the score calculation module of one embodiment, the initial score value preset for each VM is specified according to the user's service index and is in the range of 100 to 500.

To achieve the above-mentioned purpose of the invention, the embodiment provides a cloud network resource elastic scheduling method based on an integral algorithm, the method adopts the above-mentioned cloud network resource elastic scheduling system, and the scheduling method includes the following steps:

Step 1, using the virtual switch module to provide network services for each VM according to the CPU cycle ratio;

Step 2, using the data acquisition module to collect the actual number of CPU clock cycles and cycle proportion of the virtual switch consumed by each VM in each working time slice from the virtual switch module, and transmit it to the integral calculation module;

Step 3: Update the integral value of each VM using the actual number of CPU clock cycles and cycle proportion input by the integral calculation module, and transmit the updated integral value to the resource scheduling module;

Step 4: Utilize the resource scheduling module to dynamically calculate the number of virtual switch CPU cycles and cycle proportion consumed by each VM in the next working time slice according to the input integral value, and transmit the calculations to the virtual switch module.

Compared with the prior art, the present invention has the following beneficial effects:

The integration algorithm is used to bind the proportion of VM CPU cycles to VM specifications, avoiding competition for Host resources caused by pressure from multiple concurrent indicators, and effectively improving the overall utilization of Host resources.

By dynamically adjusting the proportion of CPU cycles consumed by each VM through an integral algorithm, a unified network burst transmission capacity is achieved, and better performance isolation can be achieved in this scenario.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

FIG1 is a schematic diagram of the structure of a cloud network resource elastic scheduling system based on an integral algorithm provided in an embodiment;

2 is a schematic diagram of an integral update process and a CPU cycle ratio control process based on an integral update value provided by an embodiment;

FIG3 is a schematic diagram of a mode switching of a cloud network resource elastic scheduling system provided in an embodiment.

Detailed ways

In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

By observing the network traffic pattern of cloud data centers and based on the scheduling guiding ideology of "overselling during idle times and underselling during busy times", an embodiment provides a cloud network resource elastic scheduling system and method based on an integral algorithm. By introducing an integral calculation mechanism, corresponding integrals of VM bandwidth usage and virtual switch CPU consumption are constructed, and virtual switch network resources in the next stage are reasonably allocated to ensure that the cloud network load on the entire data center tends to a reasonable level.

Fig. 1 is a schematic diagram of the structure of the cloud network resource elastic scheduling system based on the integral algorithm provided in the embodiment. As shown in Fig. 1, the cloud network resource elastic scheduling system based on the integral algorithm provided in the embodiment includes a virtual switch module, a data acquisition module, an integral calculation module and a resource scheduling module, and the bandwidth resources of the virtual logical server (VM) on the physical server (Host) are dynamically allocated through the synergy of these modules.

Among them, the virtual switch module is the core module that carries the VM network function, and is used to provide network services of corresponding specifications for the VM on the Host according to the CPU cycle ratio. Specifically, the virtual switch module includes a virtual network card interface module and a Netframe forwarding module. The virtual network card interface module is used to provide an interface for network data communication with the VM; the Netframe forwarding module is a user-mode network protocol stack component based on DPDK. As the core module for data forwarding, it is used to implement two-layer MAC address forwarding, that is, it is provided to the virtual network card interface module in the form of Netfilter Hook. The virtual network card interface module and the Netframe forwarding module both belong to the software architecture.

The data collection module is used to collect the actual number of CPU clock cycles and cycle proportions of the virtual switch consumed by each VM in each working time slice from the Netframe forwarding module included in the virtual switch module, and transmit it to the integral calculation module. Among them, the working time slice refers to a period of working time, which can be 1 second.

The integral calculation module is used to update the integral value of each VM according to the input actual CPU clock cycle number and cycle proportion, and transmit the updated integral value to the resource scheduling module. Specifically, the process of updating the integral value of each VM includes:

The integral parameters BASE, MAX and MIN are preset for each VM, where BASE is the basic cycle ratio of the virtual switch CPU consumed by the VM under the default service quality. For example, BASE can be set to 0.1, and the BASE integral at this time corresponds to 1Gbps; MAX is the maximum cycle ratio of the virtual switch CPU consumed by the VM during burst transmission. For example, MAX can be set to 0.3, and the MAX integral at this time corresponds to 3Gbps, which represents the upper limit of the network performance of the VM when the integral is available; MIN is the minimum cycle ratio of the virtual switch CPU that the VM can consume under high load or when the integral is exhausted. For example, MIN can be set to 0.02, corresponding to 200Mbps, which represents the minimum service quality of the VM. It should be noted that the specific value of BASE is related to the hardware specifications of the specific physical server and needs to be calibrated after benchmarking.

The integral calculation module also maintains the integral of each VM served. The initial integral value is set as INIT, which is set according to the user's service indicators and is generally 100 to 500. As shown in Figure 2, for each VM, the CPU cycle ratio consumed is compared to see if it is less than the BASE value. When the CPU cycle ratio is less than the BASE value, the integral is increased by X1; when the CPU cycle ratio is greater than the BASE value, the integral is reduced by X2 to obtain the integral update result, where X1 and X2 are the preset increase and decrease ratios, and where X1 is less than X2, the value of X1 is generally set to 0.1, and the value of X2 is set to 1.0, indicating that the accumulation speed of integral is slower than the consumption speed of integral.

It should also be noted that in the integral calculation module, when the original integral value of each working time slice is 0, no processing is performed and it remains 0.

The resource scheduling module is used to dynamically limit the number of virtual switch CPU cycles and cycle proportion consumed by each VM in the next working time slice according to the input integral value, as shown in Figure 2. The specific process includes: if the input integral value is 0, the maximum CPU cycle proportion consumed by the VM in the next second is set to BASE; if the integral value is greater than 0, the maximum CPU cycle proportion consumed by the VM in the next second is set to MAX; at the same time, it must be ensured that the minimum CPU cycle proportion consumed by the VM in the next second is MIN. The number of virtual switch CPU cycles and cycle proportion consumed by each VM in the next working time slice under dynamic control are transmitted to the virtual switch module, and the virtual switch module provides network services of corresponding specifications for the VM according to the received CPU cycle proportion.

The embodiment also provides a cloud network resource elastic scheduling method using the above cloud network resource elastic scheduling system, including the following steps:

Step 1: Use the virtual switch module to provide network services for each VM based on the CPU cycle ratio.

In the embodiment, after the program is loaded, each module completes parameter initialization, sets the points corresponding to each VM to the initial value, starts executing the program process, and provides network services for the VM.

Step 2: Use the data collection module to collect the actual number of CPU clock cycles and cycle proportion of the virtual switch consumed by each VM in each working time slice from the virtual switch module, and transmit it to the integral calculation module.

In the embodiment, at the end of a working time slice (one second), a query request is sent to the Netframe forwarding module; the Netframe forwarding module counts the number of CPU cycles consumed by each VM in the last second, and counts the total number of CPU cycles (including idle CPU cycles) in the past second, and returns the result to the data acquisition module.

Step 3: Use the actual number of CPU clock cycles and cycle proportion input by the integral calculation module to update the integral value of each VM, and transmit the updated integral value to the resource scheduling module.

In the embodiment, integral parameters BASE, MAX and MIN are preset for each VM. For each VM, it is compared whether the proportion of consumed CPU cycles is less than the BASE value. When the proportion of CPU cycles is less than the BASE value, the integral is increased by X1; when the proportion of CPU cycles is greater than the BASE value, the integral is reduced by X2 to obtain an integral update result, wherein X1 and X2 are preset increase ratios and decrease ratios, wherein X1 is less than X2, indicating that the accumulation speed of integrals is slower than the consumption speed of integrals.

Step 4: Utilize the resource scheduling module to dynamically calculate the number of virtual switch CPU cycles and cycle proportion consumed by each VM in the next working time slice according to the input integral value, and transmit the calculations to the virtual switch module.

In the embodiment, the points corresponding to each VM are received. If the input integral value is 0, the maximum CPU cycle ratio consumed by the VM in the next second is set to BASE; if the integral value is greater than 0, the maximum CPU cycle ratio consumed by the VM in the next second is set to MAX; at the same time, it must be ensured that the minimum CPU cycle ratio consumed by the VM in the next second is MIN.

As shown in Figure 3, the cloud network resource elastic scheduling method based on the integral algorithm and the system design scheduling scenarios: completely light load, partially high load, and completely high load.

(I) Completely light load

This condition corresponds to state 1 in Figure 3. Under this condition, the actual network overhead of all VMs reaches the BASE value, that is, the network rate of each VM is maintained at around the preset value. After a period of time, each VM has accumulated a certain number of points, and the VM can obtain network resources exceeding the limit of its BASE (but not exceeding MAX). At this time, the entire system will tolerate moderate burst transmission and high-speed flow to improve the overall resource utilization of the system.

(II) Partial high load

This condition corresponds to state 2 in Figure 3. Under this condition, the actual network overhead of some VMs reaches the BASE value, while another part of the VMs are performing burst transmission or high-speed flow, and the total network capacity of all VMs is less than or close to the total capacity of the Host. For VMs that are performing burst transmission and whose actual network overhead is greater than BASE, the previously accumulated points will be consumed. Therefore, within a certain period of time, its network rate will be limited by the virtual switch to near the BASE value, realizing the speed limit suppression function. At the same time, since the points correspond to the computing resources (CPU cycles) of the virtual switch, all VMs can obtain network resources proportional to the points, so the VMs can still maintain performance isolation.

(III) Fully high load

This condition corresponds to state 3 in Figure 3. In this condition, the total network capacity of all VMs is higher than the total capacity of the Host, so network resource competition between VMs is inevitable. At this time, the integral algorithm is no longer applicable, and the virtual switch will automatically allocate network resources according to the MIN value corresponding to each VM to ensure the availability of VM network services. If the network hotspot disappears after a period of time, the system will automatically fall back to the above two conditions; if the system still has network hotspots after a period of time, it will actively migrate some VMs on the Host to other Hosts to avoid CPU overheating and network failures.

The specific implementation methods described above have described in detail the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims (6)

1. A cloud network resource elastic scheduling system based on an integral algorithm, characterized by comprising a virtual switch module, a data acquisition module, an integral calculation module and a resource scheduling module; The virtual switch module is used as a core module that carries VM network functions and is used to provide network services for VMs according to the CPU cycle ratio; The data collection module is used to collect the actual number of CPU clock cycles and cycle proportion of the virtual switch consumed by each VM in each working time slice from the virtual switch module, and transmit it to the integral calculation module; The integral calculation module is used to update the integral value of each VM according to the input actual number of CPU clock cycles and the cycle proportion, and transmit the updated integral value to the resource scheduling module; wherein, the process of updating the integral value of each VM includes: presetting integral parameters BASE, MAX and MIN for each VM, wherein BASE is the basic consumption CPU cycle proportion, and MAX and MIN are the maximum and minimum consumption CPU cycle proportions respectively; for each VM, comparing whether the consumed CPU cycle proportion is less than the BASE value, when the CPU cycle proportion is less than the BASE value, the integral is increased by X1; when the CPU cycle proportion is greater than the BASE value, the integral is reduced by X2, so as to obtain the integral update result, wherein X1 and X2 are the preset increase ratio and decrease ratio, wherein X1 is less than X2, indicating that the accumulation speed of the integral is slower than the consumption speed of the integral; The resource scheduling module is used to dynamically limit the number of virtual switch CPU cycles and cycle proportion consumed by each VM in the next working time slice according to the input integral value to realize resource allocation, including: if the input integral value is 0, the maximum CPU cycle proportion consumed by the VM in the next second is set to BASE; if the integral value is greater than 0, the maximum CPU proportion consumed by the VM in the next second is set to MAX; at the same time, it must be ensured that the minimum CPU cycle proportion consumed by the VM in the next second is MIN; The resource scheduling module dynamically controls the virtual switch CPU cycle ratio consumed by each VM according to the input integral value and transmits it to the virtual switch module; the virtual switch module provides network services of corresponding specifications for the VM according to the received CPU cycle ratio. 2. The cloud network resource elastic scheduling system based on the integral algorithm according to claim 1, characterized in that the virtual switch module includes a virtual network card interface module and a Netframe forwarding module; The virtual network card interface module is used to provide an interface for network data communication with the VM; The Netframe forwarding module is a user-mode network protocol stack component based on DPDK. As a core module for data forwarding, it is used to implement two-layer MAC address forwarding, that is, it is provided to the virtual network card interface module in the form of Netfilter Hook. 3. According to the cloud network resource elastic scheduling system based on the integral algorithm according to claim 2, it is characterized in that the data acquisition module collects the number of virtual switch CPU clock cycles consumed by each VM in each working time slice and the proportion of the total number of cycles from the Netframe forwarding module. 4. According to the cloud network resource elastic scheduling system based on the integral algorithm according to claim 1, it is characterized in that in the integral calculation module, when the integral value is 0, the integral is not updated and the integral value is kept at 0. 5. According to the cloud network resource elastic scheduling system based on the integral algorithm according to claim 1, it is characterized in that in the integral calculation module, the initial integral value preset for each VM is specified according to the user's service indicator and the value is 100 to 500. 6. A cloud network resource elastic scheduling method based on an integral algorithm, characterized in that the method adopts the cloud network resource elastic scheduling system according to any one of claims 1 to 5, and the scheduling method comprises the following steps: Step 1, using the virtual switch module to provide network services for each VM according to the CPU cycle ratio; Step 2, using the data acquisition module to collect the actual number of CPU clock cycles and cycle proportion of the virtual switch consumed by each VM in each working time slice from the virtual switch module, and transmit it to the integral calculation module; Step 3: Update the integral value of each VM using the actual number of CPU clock cycles and cycle proportion input by the integral calculation module, and transmit the updated integral value to the resource scheduling module; Step 4: Utilize the resource scheduling module to dynamically calculate the number of virtual switch CPU cycles and cycle proportion consumed by each VM in the next working time slice according to the input integral value, and transmit the calculations to the virtual switch module.