JPH0415858A - System constitution alteration control system - Google Patents

Abstract

PURPOSE:To reduce the probability of detecting the fault of hardware by providing circuits which temporarily hold the content of a system constitution control alteration instruction from respective clusters and holding time supervisory circuits. CONSTITUTION:The circuits which temporarily hold the system constitution alteration instruction and the circuits measuring holding time are provided in system storage devices 301 and 302. When the clusters 303-306 being the objects of detachment are not in the middle of executing an operation with the other clusters, the content of a constitution control register is rewritten unconditionally. When the clusters being the objects of detachment are in the middle of executing the operation, the constitution alteration instruction is temporarily held. When the operation terminates or when holding time exceeds a prescribed value, the constitution control register is rewritten. Even if one cluster is detached by the system constitution alteration instruction in the middle of the operation such as communication between two clusters, the processing can be continued without detecting the fault of hardware.

Description

[Detailed Description of the Invention] "Summary" System configuration change control method for an entire system in which multiple clusters share a system storage device Conventionally, when two clusters in a system are communicating, one cluster is communicating with another cluster. When disconnecting one of the clusters, a system configuration change instruction is sent to the system storage device in order to solve the problem of the system stopping due to a machine check occurring due to the disconnection of the interface in the other cluster during communication. A circuit for temporarily suspending and a circuit for measuring the suspension time is provided, and if the cluster to be disconnected is not executing an operation with another cluster, the contents of the configuration control register are unconditionally rewritten, and the cluster to be disconnected is The configuration is provided with means for temporarily suspending the configuration change instruction when the operation is being executed, and rewriting the configuration control register when the operation ends or when the suspension time exceeds a specified value.

[Field of Industrial Application] The present invention relates to a system configuration change control method in an overall system in which a plurality of clusters share a system storage device, and in particular, the present invention relates to a system configuration change control method in an overall system in which a plurality of clusters share a system storage device. The present invention relates to a system change control method that allows processing to continue without detecting a machine check when a machine is disconnected from the system.

[Prior Art] FIG. 3 is a diagram showing the configuration of a system that shares a system storage device. 302 is a system storage device
(S S U : SYSTEM 5TORAGE U
NIT), 303, 304. 305. 306 each system connected to the system storage unit (SSU)
(also called “cluster”), 304a, and 302a are SSIGP (SYSTEM S [GNAL PROCE
SSOR) circuit, 303a, 304a.

305a and 306a are memory control units (MCI-A-MCU-D) in each cluster, 303
b.

304b, 305b, and 306b are the main storage devices (MSU(0) to MSU(3)>303
C, 304c, 305c, 306c are central processing units (CPU(0) to CPU(3)) in each cluster, 303d
, 304d, 305d, and 306d represent channel processors (CHPs) within each cluster.

In this example, four clusters 303 to 306 have two system storage devices (SSU (0), 5SU (1) 3
The entire system that shares 01.302 (SCM P:
SYSTEM C0UPLED MIILTIPRO
[:ESSOR) (7) This is an example.

This system storage unit (SSU) is a large-capacity semiconductor storage device that performs high-speed data transfer between main storage units (MSU) of multiple clusters.
S U) is the main storage unit (MSU) on the storage hierarchy.
And direct access memory bag! ! (It is located between DASD>.

Further, each cluster in SCMP is independent, and if a failure occurs in one cluster, another cluster on standby takes over, so that the system does not go down.

Furthermore, in SCMP, the 5srcp circuit allows
It is configured to enable inter-cluster communication via the system storage unit (SSU), and one cluster controls another cluster, and the system configuration can be changed using commands related to configuration control within the cluster. It looks like this.

[Problems to be Solved by the Invention] In the SCMP described above, a function (referred to as "high-speed hot standby") that quickly switches a cluster to a standby cluster when a failure occurs in a cluster in the system is essential. Therefore, it is necessary to change the system configuration quickly without stopping the system.

However, if a system configuration change causes the transmitting cluster to be disconnected during inter-cluster communication, a machine check will occur in the receiving cluster and the system will stop.

For example, assume that there are cluster A, cluster B, and cluster C, and cluster A and cluster B are performing inter-cluster communication via a 5SIGP circuit. During this time, if cluster C detects an abnormal state in cluster B, cluster C issues a configuration change command to disconnect cluster B, and turns off the connection signal from the configuration control register (CFR) related to the cluster. ). That one ssI
The GP circuit disconnects the communication path, thereby interrupting communication in cluster A and causing a machine check.

This means that a failure in one cluster will lead to the failure of multiple clusters.

This situation will be explained in more detail using FIG. 4 above.

That is, FIG. 4 is a diagram explaining the 5SIGP circuit, and shows the 5SIGP circuit in a block diagram. 401 is the input circuit section for each cluster in the 5SIGP circuit, and 402 is the input circuit section from the configuration control register (CFR). 403 is a register, 404a, 404b, 404c are latch circuits, 405 is a synchronization circuit, 406 is a priority circuit,
407 is a synchronization circuit, 408 is an output circuit section for each cluster in the 5srcp circuit, 409. 410 represents a register file.

In FIG. 4, data sent from other clusters (signal "SS IGP-MCU-OUT" in the diagram) is received by a register 403. If the received data is a request (REQIIEST), its contents are held in latch circuits 404a, 404b, and 404c.

Since the held data is synchronized with the clock signal of each cluster, it is synchronized with the clock signal of the system storage unit (SSU) via the synchronization circuit 405.

Then, the priority circuit 406 selects one of the requests with a high priority.

The report that the request to each cluster has been accepted is again synchronized with the clock of each cluster by the synchronization circuit 407 and is sent through the output circuit section 408.

Note that the 5SIGP circuit is occupied by the cluster until the accepted request is completed or a timeout occurs due to time monitoring.

Further, communication data for inter-cluster communication is sent to the register file 409 and synchronized with the SSU clock.

Communication data is transmitted by MCU decider (TMCUDECID).
ER), and is synchronized with the clock of the other cluster by the register file 410 and transmitted.

In the 5SIGP circuit as described above, the cluster that is the target of the operation is the configuration control register (CFR).
It is valid when the connection signal from is ON.

Therefore, an instruction to change the system configuration in the middle of a 5SIGP operation may cause a machine check in the communicating cluster, resulting in a system stoppage.

The present invention has been developed in view of the above-mentioned problems, and even if one cluster is disconnected due to a system configuration change instruction during an operation such as communication between two clusters, no hardware failure will be detected. The purpose of this invention is to provide a system configuration change control method that allows processing to continue.

[Means for solving the problem] According to the present invention, the above-mentioned object is achieved by the means described in the claims.

That is, the present invention provides a central processing unit, a channel processing unit that controls channel data transfer, a main memory unit, control of access to the main memory of each processing unit, and an interface between the central processing unit and the channel processing unit. An overall system in which a plurality of clusters composed of storage control devices, etc. share one or more system storage devices, and each of the clusters can communicate with each other via the system storage device, and the clusters are configured to share one or more system storage devices. The device has a configuration control register for controlling the configuration of the entire system, and each cluster is configured to control the separation of other clusters by rewriting the contents of its own configuration control register. In the configuration change control method, there is a circuit in the system storage device that temporarily holds system configuration change instructions from each cluster, and a circuit that measures the temporary hold time of system configuration change instructions and monitors whether the specified time has been exceeded. When a monitoring circuit is installed and one cluster in the system is disconnected due to a system configuration change instruction from another cluster, if the cluster to be disconnected is not currently executing an operation with another cluster, the hold time will be The first method is to rewrite the contents of the configuration control register in the system storage device without measuring the contents of the configuration control register, and the first method is to temporarily suspend the contents of the configuration change instruction if the cluster to be separated is performing an operation with another cluster. , a second means for starting the measurement of the pending time by the time monitoring circuit, and a second means for starting the measurement of the pending time by the time monitoring circuit; A third means of setting in the configuration control register, and if the operation being executed does not end even if the time measurement value exceeds the specified value, the contents of the temporarily suspended configuration change instruction are configured at that point. This system configuration change control method is provided with a fourth means for setting in a register.

[Operation] In the system configuration change control method of the present invention, for example, during an operation such as communication between two classes in the system, a configuration change instruction from a failure detected by another cluster other than those classes may cause the communication to be interrupted. The purpose is to prevent a machine check (considered a "hardware failure") from occurring even if one of the clusters becomes disconnected from the system.

In addition, when changing the system configuration by rewriting the contents of the configuration control register in the system storage device, (1) a circuit that temporarily holds the contents of the system configuration control change instruction from each cluster; (2) a hold time; A time monitoring circuit is installed to measure and monitor whether or not a specified time has been exceeded, and when a cluster in the system is disconnected due to a system configuration change instruction from another cluster, the cluster to be disconnected is If an operation is not being executed, the contents of the configuration control register are updated without measuring time.

If the operation is being executed, the content of the configuration change instruction is temporarily held, and the time monitoring circuit starts measuring one hour.

If the current operation ends before the time measurement reaches a certain value, the contents of the temporarily held configuration change instruction are set in the configuration control register until the next operation is started. finish.

If the operation being executed does not end even if the time measurement exceeds a certain value, it is determined that the operation cannot be completed normally, and at that point, the configuration control register is similarly updated.

In this manner, when changing the system configuration, the probability of hardware failure detection can be lowered by performing the system configuration change as much as possible during the idle time of operations between clusters.

[Example code] An embodiment of the present invention will be described below.

FIG. 1 is a diagram showing an embodiment of the present invention, and 101 is a register (CFR(1)) that holds an external configuration change instruction.
, 102 is a latch circuit, 103 is a configuration control register (
CFR (2)), 104 is a data comparison circuit (c),
105 represents a counter circuit for time monitoring.

That is, the latch circuit 102 is a circuit that temporarily holds the configuration change instruction from the register 101, and the contents of the latch circuit 102 are transmitted to the configuration control register 103. The comparison circuit 104 is a circuit that compares the contents of the latch circuit 102 and the configuration control register 103, and is a circuit that compares the contents of the latch circuit 102 and the configuration control register 103.
05 is a time monitoring circuit.

5SIGP operations are performed according to the contents of configuration control register 103.

If a configuration change command is issued during execution of a 5SIGF operation, the contents of the command are temporarily held in the latch circuit 102.

The signal a from the data comparison circuit 104 is transmitted to the latch circuit 10
When the contents of the configuration control register 103 and the contents of the configuration control register 103 are different, the flag turns ON, and the next 5SIGP operation to be executed is put on hold.

The counter circuit 105 starts time monitoring when the signal from the data comparison circuit 104 is ON.

The signal S from the counter circuit 105 is turned ON when a certain period of time has passed since the start of time monitoring.

When this signal is turned ON, the contents of the configuration control register 103 are rewritten to the contents of the latch circuit 102 even during the 5SIGP operation.

The above operation will be explained in more detail with reference to the time chart shown in FIG. 2 which explains the operation of the embodiment.

That is, in the case of a configuration change during the operation shown in FIG. 2(a), if the configuration change instruction sent by the register 101 overlaps with an operation such as inter-cluster communication, the signal "+0PERATINOBUSY" is output.
is ON, the changed contents are only held in the previous stage latch circuit 102 (signal “+○NLIN
The solid line part of “E CLUSTERPRE” is latch circuit 1.
02 indicates the data before the change, and the dashed line indicates a state where a new change instruction is held), the changed content is not received by the subsequent latch, that is, the configuration control register 103.

Therefore, rewriting to configuration control register 103 is not performed at this point, and the operation continues to be executed.

Then, the signal “+CFRU” from the data comparison circuit 104 is
The start of the next operation is temporarily stopped by "NMATCH", and the contents of the configuration control register 103 are rewritten at the end of the currently executing operation (signal "+0N
LINECLUCTER” (the broken line portion indicates that the contents have been rewritten).

In addition, if the operation continues for a long time and does not end, as shown in the time chart in Figure 2 (C) when there is a configuration change during the operation and the specified time has passed, the time monitoring in Figure 1 When a certain period of time has passed, the counter circuit 105 for
” is turned on (ON), and at that point the contents of the configuration control register 103 are forcibly rewritten.

Furthermore, if the configuration change instruction does not overlap with the operation, the contents of the configuration control register 103 are rewritten at the time the configuration change instruction is issued, as shown in FIG. 2(b).

[Effects of the Invention] As explained above, according to the present invention, for example, clusters A, B, and C among a plurality of clusters constituting SCMP
In this case, when cluster A is performing an operation with cluster B, cluster C instructs a configuration change and disconnects one of the clusters, and the processing continues without detecting a hardware failure. Can be done, SC
High reliability as an MP can be achieved.

[Brief explanation of drawings]

FIG. 1 is a diagram showing an embodiment of the present invention, FIG. 2 is a diagram showing a time chart for explaining the operation of the embodiment, FIG. 3 is a diagram showing the configuration of a system that shares a system storage device, FIG. 4 is a diagram explaining the 5SIGP circuit. (MCU) , 303b, 304b, 305b,
306b... Main storage unit (MSU), 3
03c, 304c, 305c, 306cm...
・Central processing unit (CPU), 303d, 304
d, 305d. 306d...Channel processor (CHP),
401...5 Input circuit section for each cluster in the SIGP circuit, 402... Latch circuit, 403
...Register, 404a, 404b, 40
4C---=latch circuit, 405°407...
. . . Synchronous circuit, 406 . . . Priority circuit, 408 . . . Output circuit section for each cluster in the 5 SIGP circuit, 409. 410・・・Register file

Claims (1)

[Claims] A central processing unit, a channel processing unit that controls channel data transfer, a main storage unit, and a storage control unit that controls access to the main memory of each processing unit and the interface between the central processing unit and the channel processing unit. An overall system in which a plurality of clusters composed of devices, etc. share one or more system storage devices, and each of the clusters can communicate with each other via the system storage device, and the clusters in the system storage device share one or more system storage devices. has a configuration control register for controlling the configuration of the entire system, and each cluster is configured to rewrite the contents of the configuration control register to control the separation of other clusters. In the system storage device, there is a circuit that temporarily holds system configuration change instructions from each cluster, and a time monitoring circuit that measures the temporary hold time of system configuration change instructions and monitors whether the specified time has been exceeded. When one cluster in the system is detached due to a system configuration change instruction from another cluster, if the cluster to be detached is not currently executing an operation with the other cluster, the hold time is measured. The first method is to rewrite the contents of the configuration control register in the system storage device without having to do so, and the first method is to temporarily suspend the contents of the configuration change instruction if the cluster to be detached is performing an operation with another cluster, and to save time. A second means for starting the measurement of the suspension time by the monitoring circuit, and a configuration control for the contents of the temporarily suspended configuration change instruction if the operation being executed ends before the time measurement reaches a predetermined value. A third means of registering the contents of the temporarily suspended configuration change instruction in the configuration control register, if the operation being executed does not end even if the time measurement value exceeds the specified value. A system configuration change control method, characterized in that a fourth means for setting is provided.