JPH06252939A - Method for maintaining synchronization status between duplicated processors - Google Patents

Abstract

(57) [Abstract] [Purpose] To provide a synchronization state maintaining method for maintaining the synchronization state between duplicated processing devices without increasing communication / processing overhead. [Structure] In order to maintain a synchronized state between the duplicated processing devices which are connected via a communication path 50 and one constitutes an act system control unit 30 and the other constitutes a standby system control unit 31, periodically or On-demand, or periodically and on-demand, the status data representing the status of the act system control unit 30 is sent to the standby system control unit 3 via the communication path 50.
Transfer to 1.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for maintaining a synchronized state between duplicated processing devices, and more particularly to a method for maintaining a synchronized state suitable for processing devices which constitute a control unit of an ATM switch.

[0002]

2. Description of the Related Art With the advent of the advanced information society, processing devices using electronic computers have come to be widely used for real-time applications such as transaction processing and control of various information communication devices. In such an application, if the service is interrupted, the user suffers a great deal of damage. Therefore, there is a demand for a structure in which the service is not stopped even if a failure occurs. Duplication technology is an important technology for that purpose. Redundant technology has the same processing unit (electronic computer) in duplicate, one of which normally provides services as an act system and the other stands by as a standby system, but a failure occurs in the act system. In this case, the standby system becomes a new act system and takes over the service.

In the duplication of electronic computers, the intended purpose cannot be achieved by simply installing the computers in duplicate. In order for the standby system to normally take over the service when the act system fails, the standby system must take over the state of the act system. The status inheritance from the act system to the standby system is performed by transferring the status data indicating the status from the act system to the standby system. The takeover of this state is too late after a failure occurs in the act system and it is necessary to switch the system from the act system to the standby system. This is because there is no guarantee that the faulty act system can normally transfer the state data to the standby system.

Therefore, it is customary to take over the state at all times even in a normal state. That is, the redundant computers are connected by a communication path.By using this communication path, the act system constantly transfers its own state data to the standby system, and the standby system transfers the transferred state data to its own memory function. Remember. As a result, the states of the act system and the standby system are matched.

In such a duplex system, it is ideal that the states of the act system and the standby system are always the same. However, if you try to transfer state data to the standby system each time there is a change in the state data in the act system as ideal, the communication overhead and processing overhead for that purpose become a bottleneck, and the throughput of the entire computer system can be increased. Sometimes there is no.

[0006] This is the B-ISDN which has recently begun to attract attention.
(Broad-Band Integrated Services Digital Network)
ATM (Asynchronous Transfer Mod) to realize
e: Asynchronous transfer mode) This is a particular problem when the control unit of the exchange is duplicated. At the same time as the ATM switch is required to have fault tolerance, it is also required to improve the processing capability of the control unit commensurate with the size of the exchange capacity. In such a case, when the state inheritance method in the conventional duplex system is used, there is a problem that the processing capacity cannot be improved as expected due to communication / processing overhead.

[0007]

As described above, according to the conventional duplication technique, in order to always make the states of the act system and the standby system coincide with each other, the state data of the act system is changed to the standby system whenever the state data changes. However, there is a problem in that the processing capability cannot be expected to be improved due to the communication / processing overhead because the transfer method is adopted.

Therefore, an object of the present invention is to provide a synchronization state maintaining method for maintaining the synchronization state between duplicated processing devices without increasing communication / processing overhead.

[0009]

SUMMARY OF THE INVENTION In order to solve the above problems, the present invention relates to an act system which is connected through a communication path and one of which actually performs a service and the other of which is a standby system in which a service is suspended. Maintains a synchronized state between the duplicated processing units of the same structure that are set to play respective roles and set the standby system to take over the state of the act system and continue the service when the act system enters the service suspension state. In the method described above, by periodically transmitting the state data representing the state of the act system to the standby system through the communication path, periodically or in response to the request, or periodically and in response to the request, It is characterized by maintaining a synchronized state.

Further, according to the present invention, in such a basic configuration, the storage space of the storage means for storing the state data is divided into a plurality of blocks, and the state data changed after the previous transfer is stored in these blocks. It is characterized in that only the current block is rewritten, and only the state data in the rewritten block is collectively transferred from the act system to the standby system.

[0011]

As described above, according to the present invention, the transfer of the state data from the act system to the standby system is performed periodically or at any time in response to a request, or periodically and at any time in response to a request, thereby performing communication. -Processing overhead is reduced.

That is, in the conventional method of transferring the state data to the standby system each time there is a change in the data of the act system, not only the number of transfers increases, but also the destination and other control information attached to the state data when transferring the state data. Since it is necessary to transfer the data at the same time, the communication amount has increased. On the other hand, in the present invention, the amount of state data transferred at one time increases, but the number of transfers decreases, and the communication amount required to transfer additional control information decreases.

Further, the storage space of the storage means for the state data is divided into a plurality of blocks, only the blocks changed after the previous transfer are rewritten, and only the state data in the rewritten blocks is collectively transferred from the act system to the standby system. If transferred, the communication and processing overhead will be further reduced due to the further reduction in communication volume.

Further, in the conventional method, in order to reduce communication / processing overhead, if a batch transfer including changed status data and unchanged status data from the act system to the standby system is attempted, a high speed dedicated bus is used. However, if only the state data of the changed block is transferred as in the present invention, such a dedicated bus is not required.
The control path realized by M can be used to transfer the state data.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing an embodiment in which the present invention is applied to the duplication of a control unit in an ATM exchange.

In FIG. 1, an ATM switch 10 is a switch circuit that executes ATM exchange of cells, and is usually a so-called self-routing ATM switch that automatically guides the cells to an output port according to a routing tag included in the internal cells. It is composed of

The subscriber cards / trunk cards 20 to 23 connected to the ATM switch 10 perform (a) label conversion for the UNI / NNI cells from the terminal / adjacent ATM switch, and output from the ATM switch 10. (B) Remove the routing tag from the internal cell format cell output from the ATM switch 10 to convert it to the internal cell format with a routing tag that specifies the port U
It has the function of converting to the NI / NNI cell format and sending it to the terminal / adjacent ATM exchange. AT like this
The details of the M switch and the subscriber card / trunk card are not directly related to the gist of the present invention, and thus the description thereof will be omitted. For example, the configuration described in Japanese Patent Application No. 2-217216 "ATM communication system" is used. be able to.

The ATM switch 10 is connected to the redundant control units 30 and 31, one of which is an act (AC).
The T) system control unit 30 is used as the standby (SBY) system control unit 31. An ACT control unit 30,
An AT is connected to the control unit or terminal of another ATM exchange.
The control paths 40 to 43 realized by M are set.
The ACT system control unit 30 communicates with the control units and terminals of other ATM exchanges using these control paths 40 to 43, and implements call control, operation management, and the like.

A control path 50 realized by ATM is set between the ACT system control unit 30 and the SBY system control unit 31. The ACT system control unit 30 controls the control path 50.
Is used to notify the SBY system control unit 31 of the latest state by transferring the state data, and the SBY system control unit 31 updates the state to the notified latest state. As a result, the ACT system control unit 30 and the SBY system control unit 31 maintain the synchronized state.

In this embodiment, as described above, the connection between the ACT system control unit 30 and the SBY system control unit 31 is realized by the ATM connection. This greatly contributes to the realization of a low-cost ATM switch, and the ACT control unit 30
It also greatly contributes to the flexible deployment of the SBY control unit 31. In fact, since the ACT system control unit 30 and the SBY system control unit 31 are connected by the ATM connection,
Both of them may be connected to any port of the ATM switch 10.

Further, the ACT control unit 30 and SBY.
The system control unit 31 does not even have to be in the same ATM exchange. For example, as shown in FIG. 2, the SBY system control unit 31
It may be housed in the adjacent ATM exchange 11 connected to the ATM exchange 10 to which the ACT system control unit 30 is connected. However, when the control unit switches from the ACT system control unit 30 to the SBY system control unit 31, the control paths 40 to 43 with the control units / terminals of other ATM exchanges are transferred from the ACT system control unit 30 to the SBY system control unit 31. However, it is easier for the ACT system control unit 30 and the SBY system control unit 31 to be in the same node. That is, if the ACT system control unit 30 and the SBY system control unit 31 are in the same node, the reconnection of the control paths 40 to 43 may be performed only in the part in the node. This is because it is necessary to reconnect the outer part.

Next, a method for maintaining the synchronized state between the ACT system control unit 30 and the SBY system control unit 31 will be described in detail. ACT system control unit 30 and SBY system control unit 31
The configuration of is shown in FIG. As shown in the figure, the configurations of the ACT system control unit 30 and the SBY system control unit 31 are the same, and the CPUs 301 and 311 and the hard disks 302 and 3 respectively.
12, memory 303 and 313, and protocol processing units 304 and 314 for ATM communication. The storage spaces of the memories 303 and 313 are program storage areas 3031 and 3131 for storing programs, status data storage areas 3032 and 3132 for storing status data to be inherited from the ACT system control unit 30 to the SBY system control unit 31, and work data. Work data storage area 3033 for storing
It is divided into 3133.

In this embodiment, the ACT system control unit 30 uses SB
It manages and controls the synchronous operation state of the Y-system control unit 31. That is, the operating state of the SBY system control unit 31 is determined by the judgment of the ACT system control unit 30, and the process for changing the state is started. The operating states of the SBY system control unit 31 include a synchronous state (built-in state) and an isolated state (disconnected state). The synchronized state means a state (heat preliminary state) in which the program and state data of the SBY system control unit 31 are maintained the same as the program and state data of the ACT system control unit 30 (data synchronization is established). The isolated state means that the program and state data of the SBY system control unit 31 are the ACT system control unit 3.
A state in which the program and state data of 0 do not match.

The ACT system control unit 30 is replaced with the SBY system control unit 31.
The processing for controlling the state of SBY includes SBY system synchronization processing, SBY system synchronization maintenance processing, and SBY isolation processing. A schematic flow of these processes is shown in FIG.

The SBY system synchronization processing is performed by the SBY system control unit 3
1 is a process for shifting the isolated state from the isolated state to the synchronized state, and the SBY system control unit 31 loads the program from the own system hard disk 312, so that the SBY system control unit 31
After the programs of the ACT system control unit 30 become the same,
It is activated by an incorporation request from the SBY system control unit 31. In this SBY system synchronization processing, the ACT system control unit 30
And the SBY system control unit 31 all state data that must be matched is transferred from the ACT system control unit 30 to the SBY system control unit 31.

The SBY system synchronization maintaining process is a process for maintaining the SBY system control unit 31 synchronized by the SBY system synchronization process in a synchronized state, and is automatically started when the SBY system synchronization process is completed. In this SBY system synchronization maintaining process, A
When the CT system control unit 30 updates the state data, only the updated state data is notified to the SBY system control unit 31, and S
The status data of the BY system control unit 31 is updated. In the present embodiment, there are two methods for transferring the changed status data from the ACT system control unit 30 to the SBY system control unit 31: a cyclic method as described later and an on-demand method. Are offered.

The SBY system isolation process is performed by the SBY system control unit 3
1 is a process for shifting 1 from the synchronized state to the isolated state, and is achieved by stopping the SBY system synchronization maintaining process. As a result, even if the state data is updated by the ACT system control unit 30, the SBY system control unit 31 is not notified, so that the state data of the ACT system control unit 30 and the SBY system control unit 31 do not match.

In this embodiment, (1) SBY synchronous state control (2) ACT → SBY data transfer control is provided to the application program as a system call for synchronous control.

A system call for initialization may be required in addition to this, but it is omitted because it is unrelated to the essence of the present invention.

First, the SBY synchronous state control will be described. The SBY system synchronization state control is a system call group for the ACT system application to manage / control the SBY system synchronization state. Specifically, an SBY system sync state setting system call for the ACT application program to set the SBY system sync state (in sync / isolation) and an SBY system in which the ACT application refers to the current SBY system sync state Consists of a sync state reference system call. The ACT system control unit 30 has an SBY system state flag indicating the SBY system synchronization state, and the SBY system synchronization state setting system call sets the SBY system state flag. Also, the SBY system synchronization status reference system call refers to the flag to indicate the current synchronization status of the SBY system control unit 31 to ACT.
Notify system applications.

Next, the ACT → SBY system data transfer control will be described. ACT → SBY system data transfer control is
In order to match the state data between the ACT system control unit 30 and the SBY system control unit 31 during the SBY system synchronous operation, the AC
It is a system call group that provides a function of transferring status data from the T system control unit 30 to the SBY system control unit 31. The transfer of state data is done in block copy images.
That is, the transfer target range on the memory is registered as a data block of an appropriate size, and the AC
The state data is transferred from the T system control unit 30 to the SBY system control unit 31. Therefore, the ACT → SBY data transfer control provides a function of registering a data block and a function of transferring the content of the designated data block to the SBY system control unit 31.

In the transfer of the status data, the layer 2 function is used to retransmit when an error occurs. Two types of state data transfer methods are provided: a cyclic method (hereinafter, referred to as cyclic transfer) and an on-demand method, that is, a method at any time according to a request (hereinafter, referred to as on-demand transfer).

In the periodic transfer, the transfer is performed when the synchronous operation is performed, and the transfer is automatically stopped when the isolated state is entered.
In this cyclic transfer, first, which data block is transferred to the SBY system control unit 31 at which cycle is registered. FIG. 5A shows the state of this registration, and in this example, B1, B2 and B5 are transferred in cycle 1.
3 blocks, B3 and B6 are transferred in cycle 2, B4 is transferred in cycle 3, and B7 is transferred in cycle 4. Thus, by preparing a plurality of types of cycles for cyclic transfer, it is possible to perform finely-tuned synchronous operation control according to the characteristics of each data block.

The system calls that support the above cyclic transfer registration include a transfer cycle registration system call and a cyclic transfer request system call. The transfer cycle registration system call is for registering the correspondence between the transfer cycle and the ID number. For example, the transfer cycle of cycle 1 is registered as transfer cycle ID = 0. The cyclic transfer request system call is for registering which ID of each data block is transferred in a transfer cycle, and sets one data block that needs to be transferred.

FIG. 5B is an example showing a state of periodic transfer in time series. However, in this figure, it is assumed that two cycles, cycle 1 and cycle 2 (assumed to be twice cycle 1), have already been registered by the transfer cycle registration system call. When the transfer cycle is registered, the timer of the registered cycle is started regardless of the presence or absence of the data block to be transferred by the transfer cycle.

First, attention is paid to the operation on the cycle 1 side. B1,
When B2 is registered, as shown in FIG. 5B, the cyclic transfer is started from the time when the cycle of registration is completed. However, in the present embodiment, in order to reduce the transfer data amount as much as possible, only the data blocks changed during each cycle are transferred. That is, both B1 and B2 are transferred at the end of the first cycle, but only B1 is changed at the next cycle, so only B1 is transferred at the end of the cycle and both B1 and B2 are changed at the next cycle. Therefore, neither B1 nor B2 is transferred at the end of the cycle. After that, when B5 is registered, the cyclic transfer is similarly started from the end of the cycle in which the registration is performed. On the other hand, the same applies to the cycle 2 side,
When B3 and B6 are registered, the cyclic transfer of the data block is started at the end of the cycle. However, as described above, the transfer at the end of the cycle is omitted for the data block in which the state data does not change within the cycle.

In the periodic transfer, when system switching occurs during status data transfer, the new ACT system control unit (SBY system control unit before system switching) is rewriting the state data on the memory, so the data contents May be inconsistent. This detects and initializes inconsistencies during the startup process,
It can be dealt with by performing a logical check. Therefore, the cyclic transfer control is intended for data that can be logically checked.

FIG. 6 shows a schematic flow of processing contents at the time of periodic transfer. In the ACT control unit 30, first, the application program registers the transfer cycle as described above (S10). This activates the timer in the periodic block copy task 305.

Next, when the application program registers the data block to be transferred (S11), the periodic block copy task 305 transfers the data block registered in the period corresponding to the timer when the timer interrupt is input. However, when the SBY system synchronization status flag 307 indicates that the SBY system synchronization status is isolated, the transfer process is not performed. In addition, the transfer process is suppressed for the data block that has not changed from one cycle before. The data block transfer uses the layer 2 function 308 in order to perform a retransmission process when an error occurs. In FIG. 6, the memory 303
Since the data blocks # 0 and # 1 in 2 have changed within the cycle, the transfer process is being performed.

The SBY system control unit 31 performs an error check on the transferred data block using the layer 2 function 318, and if it is normal, the write task 316 writes the data block to the memory 3132. If there is an error, the ACT system control unit 30 is requested to retransmit.

Next, the on-demand transfer will be described. On-demand transfer is a method of transferring state data when the application issues a transfer request.
It is designed so that the transfer method itself has a mechanism for guaranteeing the consistency of data contents. That is, the ACT system control unit 30
When the (sending side) starts updating a certain data block,
Writing to the data block is prohibited until the transfer after the update is completed. Therefore, the OPEN system call that puts the data block in a state where access is prohibited from others,
A CLOSE system call for activating the copy of the data block and releasing the access prohibited state when the copy is completed is provided.

The OPEN system call uses the list of data blocks to be transferred, the request type (immediate request / waiting time request), and the access type (WRITE / READ) as input arguments to make an OPEN request for the requested data block. , The request identifier is returned as an output argument. OPEN of READ is permitted for a data block that has already been OPENed, but WRIT
OPEN of E is not allowed. If OPEN is not allowed, an immediate error is returned if the request is an immediate request, but if the request is a wait request, wait until OPEN is allowed. OPE
N system calls are processed one at a time. That is, when a waiting request is generated during the processing of the previous system call, the waiting request after that is queued and waits. On the other hand, the CLOSE system call uses the request ID given by the OPEN system call as an input argument to C.
Request LOSE.

An application program which wants to access the data block must always make an OP immediately before the access.
Make an EN call to check if access is possible. In this way, the contents of the memory in the ACT system control unit 30 are not rewritten by others until the transfer is completed.
The consistency of the data content transferred from the T-system control unit 30 is guaranteed. However, a process of trying to access the data block in the exclusive state causes a refusal / wait.

Even if the above-mentioned measures alone can guarantee the consistency of the state data transferred from the ACT system control unit 30, the system is switched during the data transfer from the ACT system control unit 30 to the SBY system control unit 31. Therefore, the consistency of data in the SBY system control unit 31 cannot always be guaranteed.

Therefore, in the present embodiment, it is stored that the data is being rewritten from the start of the transfer to the SBY system control unit 31 (reception side) to the end of the rewriting of the memory contents, and the restart processing after the system switching is performed. A system call is provided to refer to this data rewrite state from the inside. As a result, the new ACT system control unit can know whether system switching has occurred during the state data transfer. This on-demand transfer is used for transferring important state data, which is difficult for the new ACT control unit 31 to detect data inconsistency or initialize during startup processing.

FIG. 7 and FIG. 8 show a schematic flow of processing during on-demand transfer. In the ACT control unit 30, first, as shown in FIG. 7, the application program calls the OPEN system call to prohibit access to the designated data block (S2).
0). Next, as shown in FIG. 8, processing such as rewriting of the OPENed data block is performed. When the processing for this data block is completed, the application program calls the CLOSE system call,
The specified data block (data blocks # 0, # 1 in the figure) in the memory 3032 is transferred to the SBY system control unit 31.
Then, the access right to the data block is abandoned (S21).

The state data transfer to the SBY system control unit 31 is performed only when the SBY synchronization state flag 307 indicates that synchronization is in progress, and is not performed when it is in isolation. The state data transfer also uses the layer 2 function 308 to perform a resend process if an error occurs. SB
The Y system control unit 31 sets a data rewriting flag 319 when the state data transfer from the ACT system control unit 30 starts, and clears the flag 319 when the data transfer ends. The data rewriting flag 319 can be referred to from the application program (S22). Thereby, the SBY system control unit 31 can confirm the consistency of the state data.

FIG. 8 shows OPEN of READ during data transfer.
The call is allowed, but the WRITE OPEN call is not allowed. WRITE OPEN
The call is allowed only after the transfer process is completed.

As described above, periodical and on-demand transfer of state data, that is, memory copy can be realized. In this method, particularly when applied to a duplicated control unit of an ATM switch, a conventional method of immediately transmitting the state data to the SBY system control unit and notifying the state whenever the memory content (state data) is changed Compared with (1) same A
(2) CP for ATM communication that can minimize the effect on user traffic accommodated in the TM switch
The advantage is that the U overhead can be minimized.

The present invention is not limited to the duplication of the control unit of the ATM switch, but the duplication of the control unit of other information communication equipment,
It goes without saying that it can be widely applied to the duplication of electronic computers that perform online processing.

[0051]

As described above, according to the present invention, it is possible to provide a method for maintaining a synchronized state between duplexed processing devices with less communication overhead and processing overhead. INDUSTRIAL APPLICABILITY The method of the present invention is particularly suitable for duplication of the control unit of an ATM exchange, and it is possible to provide an ATM exchange that is resistant to failures and has a small decrease in throughput when duplicated.

[Brief description of drawings]

FIG. 1 is a diagram showing an embodiment in which the present invention is applied to duplication of an ATM exchange control unit.

FIG. 2 is a diagram showing an embodiment when an act system control unit and a standby system control unit exist in different nodes.

FIG. 3 is a configuration diagram of an act system control unit and a standby system control unit.

FIG. 4 is a schematic flowchart of an act system control unit and a standby system control unit.

FIG. 5 is a diagram showing a transfer cycle registration example of status data and a time-series status at the time of cyclic transfer.

FIG. 6 is a schematic flow chart at the time of periodic transfer of status data.

FIG. 7 is a schematic flowchart of on-demand transfer of status data.

FIG. 8 is a schematic flowchart of on-demand transfer of status data.

[Explanation of symbols]

10, 11 ... ATM switch 20-23 ... Subscriber / trunk card 30 ... Act system control unit 31 ... Standby system control unit 40-43 ... Control path 50 ... Control path 301, 311 ... CPU 302, 312 ...
Hard disk 303, 313 ... Memory 304, 314 ...
Protocol processing unit 305 ... Periodic block copy task 307 ... SBY synchronization status flag 308, 318 ... Layer 2 function 316 ... Write task 319 ... Data rewriting flag 3031, 3131 ... Program storage area 3032, 3132 ... Inherited data storage area 3033, 3133 … Work data storage area

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Ryuji Ito 1 Komukai-shi Toshiba-cho, Kawasaki-shi, Kanagawa Prefecture Corporate Research & Development Center, Toshiba Corporation (72) Inventor Takashi Kamitake Komukai, Kawasaki-shi, Kanagawa Toshiba Town No. 1 Inside Toshiba Research and Development Center

Claims (2)

[Claims] 1. When the act system is connected through a communication path, one of which acts as an act system that actually performs a service, and the other of which serves as a standby system in which the service is suspended, and the act system enters a service suspension state. In the method for maintaining the synchronized state between the duplicated processing devices of the same structure set so that the standby system takes over the state of the act system and continues the service, periodically or at any time when requested. Of the act system periodically, or periodically as needed, by transferring state data representing the state of the act system to the standby system through the communication path, thereby maintaining the synchronous state. For maintaining a synchronized state between duplicated processing devices. 2. A storage means for storing the status data is provided, the storage space of the storage means is divided into a plurality of blocks, and the status data that has been changed after the previous transfer is stored in these blocks. Rewrite only the block,
2. The method for maintaining the synchronized state between the duplicated processing devices according to claim 1, wherein only the state data in the rewritten block is collectively transferred from the act system to the standby system.