JP2856761B2 - Resource lock management device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Field of Industrial Application) The present invention relates to a resource lock management device for a computer system.

(Prior Art) Many of modern computer systems perform multiplex processing, and tasks of a plurality of users often run on the computer at the same time. In addition, there are many unspecified users who use these computer systems in many cases, and it cannot be expected that the use of computers will be mutually strictly agreed.

On the other hand, the computational resources such as data required to perform work on these computers are becoming increasingly enormous with the progress of computers, and it is necessary to secure large resources individually for each user. It is difficult, and it is convenient to put together what you can share. In addition, by doing so, the data can be updated at one time, and it is easy to maintain the consistency of the data used by the user.

However, in the case of sharing resources such as data, it is known that inconsistency occurs in data if a plurality of users are allowed to update data randomly. For this reason, the process that executes the work of each user locks the resources required by itself, and eliminates the process of accessing data by another user until the necessary processing is completed. Has been taken.

On the other hand, in a computer system, a certain set of data is in the form of one file. Each process reads from and writes to the parts that it needs from this file while working on it. Usually, data that is shared by a plurality of users is large in size, and the file in which it is stored is also large. Thus, the likelihood of multiple processes using a portion of the file at the same time is not very large. In such a situation,
If a lock for exclusion is set for each file, if two processes try to access different parts of the file, one of the processes will wait for execution even though there is essentially no conflict. become.

In order to avoid such absurdity, a file is divided into smaller units, and locking is performed using the divided units.

In addition, in order to realize a lock on a resource shared by a plurality of users including such data,
Usually, a resource lock management table LT as shown in FIG. 4 is used. The lock management table has as many entries as the number of resources to be locked. Each entry has a lock flag indicating whether the resource is already locked, and a process number field storing the process number of the process locking the resource. When a certain process A reads / writes from / to a certain resource X, the process A first checks an entry corresponding to the resource X in the lock management table. If the lock flag of the entry is not set, that is, if the resource X has not been locked by another process, the resource X can be acquired, and the process A sets the lock flag and puts its own process number in a predetermined field. Register and update the lock management table LT. If the flag is set, various methods have been proposed as to what to do with the processing. In this case, a state in which process A temporarily abandons acquisition and waits for resource X to be released (block State).

However, if a plurality of processes running on different processors are allowed to access the lock management table LT at the same time, a deadlock may occur as described below.

That is, it is assumed that two processes A and B are being executed, and both are trying to acquire two resources of resource numbers 1 and 2. Now, it is assumed that the process A attempts to acquire the resource 1 and updates an entry (entry 1) corresponding to the resource 1 in the lock management table. Next, it is assumed that the process B updates an entry (entry 2) corresponding to the resource 2 in the lock management table LT before the process A attempts to acquire the resource 2. Then, process A goes to check entry 2 in an attempt to acquire resource 2, but because process B has already locked it, process A is blocked from executing. Process B also attempts to acquire resource 1 but is already locked by process A, so process B is also blocked.

As described above, a situation in which the execution is stopped waiting for the processes A and B to release resources from each other, that is, a deadlock occurs. This deadlock is
This is because processes A and B are allowed to access the resource lock management table LT at the same time.

In order to avoid the above situation, it is conceivable to completely and exclusively control access to the lock management table LT. In this case, while one process is accessing the lock management table, the other processes cannot access the lock management table during that time, and thus have to wait for execution.

(Problem to be Solved by the Invention) However, in the conventional method in which the other is completely exclusive, since all processes that attempt to access a certain resource occupy one lock management table, contention there is caused. Could adversely affect system performance.

The present invention has been made in view of such circumstances, and an object thereof is to adjust contention for access to a lock management table and improve the performance of a computer system.

[Configuration of the Invention] (Means for Solving the Problems) In order to achieve the above object, a resource lock management device of the present invention has a plurality of resources shared by a plurality of processes executed by a plurality of processors. A resource lock management device that locks a specific resource with respect to another process and then occupies the specific process with a specific process, comprising: a plurality of lock management tables provided in a number smaller than the number of the plurality of processes; Means for selecting and allocating an unused lock management table among the plurality of lock management tables, detecting which resource each process is requesting, and based on the detection result, Process that has made an access request for a process and a process that has made an access request that does not interfere with the access requests of other processes. Means for permitting the locking Te, and has a means for updating simultaneously for locking results by means for permitting the locking process other lock management table.

(Operation) According to the resource lock management device of the present invention, the means for simultaneously updating the lock result has a plurality of lock management tables which always reflect the current lock result, and the allocation means responds to the access request from the process. Instead of allocating a specific management table, instead of allocating a lock management table that is not used among multiple lock management tables, it is possible to speed up the processing when performing lock processing by multiple processes in parallel. Can be achieved.

In particular, since the lock process is permitted by the means for permitting the lock process for an access request that does not interfere with the earliest access request and the access request of another process, the access requests that interfere with each other are set. By eliminating the lock process, it is possible to prevent inconsistency in the lock processing.

At this time, the means for permitting the lock process monitors the content of the access request from each process by detecting which resource each process is requesting, and ensures that the access requests that interfere with each other are detected. Can be eliminated.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of a resource lock management device according to the present invention. In the figure, 1a, 1b... 1n indicate processors.

Each processor 1a, 1b... 1n has a data bus (2a, 2b.
Address bus (3a, 3b ... 3n), read / write identification signal line (4
a, 4b ... 4n), lock management device access signal lines (5a, 5b ... 5)
n) and connected to the access arbitration device 8 of the lock management device 7 via control signal lines such as approval signal lines (6a, 6b... 6n).

The lock management device 7 has a plurality (two in this example) of copy storage devices 9 and 10 for recording copies of the lock management table LT shown in the conventional example. The maximum number of copy storage devices may be the same as the number of processors, but may be appropriately adjusted according to the frequency of access collision of the management table.

The processors 1a, 1b... 1n are connected to the copy storage devices 9, 10 via the channels (11i, 11j) in the access arbitration device 8. Each copy storage device 9, 10 is controlled via control signal lines such as an internal address bus (12i, 12j), an internal data bus (13i, 13j), and an internal read / write identification signal line (14i, 14j). It is connected to the monitoring device 15. By monitoring the internal address buses (12i, 12j), the monitoring device 15 can know which entry is being accessed, that is, which resource is being requested.

In the access arbitration device 8, acknowledgment signal lines (6a, 6b... 6n) are connected to the processors 1a to 1n. This acknowledgment signal line corresponds to a data ready signal in normal memory access. If this signal is ON, the processor judges that the memory is ready for reading and writing and executes a read / write cycle. . If this signal is OFF, the read / write cycle is waited. In this way, the execution of the processors 1a to 1n can be controlled.

As shown in detail in FIG. 2, the access arbitration device 8
A status flag indicating whether each of the processors 1a to 1n is accessing the lock management device 7 or not.
16 corresponding to each of the processors 1a to 1n. Also,
The access arbitration device 8 has a busy flag 17 indicating whether a copy of each management table LT is allocated to one of the processors, corresponding to each of the copy storage devices 9 and 10. Initially, all of these flags are in the reset state.

As shown in FIG. 3, the processing from the start of access to the lock management device 7 to the end of the process includes a read phase F1, an update phase F2, and an end phase F3.
And can be divided into

First, assuming that there is no contention for access to the lock management device 7, the operation of the process in each phase and the operation of the corresponding lock management device 7 will be described.

Now, a certain process A running on the processor 1a
Assuming that the processor 1a wants to access the lock management device 7, in the reading phase F1 of the process A, the processor 1a turns on the control signal from the access signal line 5a of the lock management device 7 and simultaneously needs the address bus 3a. The address corresponding to the entry in the management table of the resource to be used is placed. At this time, the access arbitration device 8 determines that this is the first access by the process A to the lock management device 7 based on the rise of the control signal on the access signal line 5a of the lock management device, and is not assigned to any processor. If there are copy devices 9 and 10, one of them (for example, copy storage device 9) is assigned to processor 1a.

Specifically, the access arbitration device 8 has its processor 1a
Is set, and the hegi flag 17 corresponding to the copy apparatus 9 is set. Then, the channel 11i from the processor 1a to the copy storage device 9 is allocated. If all copy storage devices have already been allocated to another processor, the access arbitrator 8 does not return an acknowledgment signal, and the processor 1a waits until any copy storage device is released. When the copy storage device 9 is allocated, the process A checks the required entry and determines whether the lock flag is not set and the resource can be acquired.
At this time, the monitoring device 15 monitors which resource the process A is trying to acquire. However, as long as there is no access to the lock monitoring device 7 from another process, the monitoring device 15
15 does not interfere with the execution of process A. When it is determined that all the required resources can be acquired, the process A enters the update phase F2. Otherwise, since the process A cannot acquire the resource, it skips the update phase and enters the end phase F3.

In the update phase F2, the process A updates an entry in the lock management table LT corresponding to the resource to be acquired. When the access arbitration device 8 detects that the process A has entered the update phase F2 by the read / write identification signal, the access arbitration device 8 updates all the copy storage devices 9 and 10 so that all the copy storage devices 9 and 10 can be updated simultaneously. 10 processor
Assign to 1a.

In the update phase F2, when necessary updates to the lock management table LT have been completed, or in the read phase
If it is determined in F1 that the necessary resources cannot be obtained, the process A enters the end phase F3. At this time, the processor 1a turns off the control signal of the lock management device access signal line 5a to notify that the access to the lock management device 7 has been completed. In response to this, the access arbitrating device 8 resets the status flag 16 for the processor 1a and the busy flag 17a corresponding to the copy storage device 9 assigned to the processor 1a. The monitoring device 15 clears the record of which resource the process A has requested.

Next, how the lock management device 7 executes the access in parallel when the lock management table LT is accessed by two processes will be described.

Now, a process A running on the processor 1a,
It is assumed that both processes B running on the processor 1b access the lock management device 7. Then, the access arbitrating device 8 assigns the copy storage devices 9 and 10 of the lock management table LT to each. Here, for the process A, the copy storage device 9
Assume that copy storage 10 has been allocated to process B.

The start of access to the lock management table by a new process is also transmitted to the monitoring device 15, and the monitoring mechanism is started. That is, the monitoring device 15 monitors the internal address buses 12a and 12b, records which resources are being requested, and constantly checks whether or not the resources requested by the processes A and B overlap. doing.

If the resources to be acquired by process A and the resources to be acquired by process B do not overlap, both processes operate independently. As a result, access to the lock management table LT can be multiplexed, so that performance degradation due to contention for access rights can be prevented.

If an overlap is detected, the monitoring device 15 does not return an acknowledgment signal to the processor that has accessed the entry that was accessed later, but waits.
Here, it is assumed that the processor 1b has been weighted. In this case as well, the processor 1b is kept waiting until the process A goes through the end phase F3 to end the access. In the meantime, the access arbitrating device 8 again allocates the copy storage device 10 to the processor 1b, and
Returns an approval signal to the processor 1b, and the execution of the process B is continued.

As described above, in this example, by the monitoring operation of the monitoring device 15,
If there is no interference in processes A and B, allow parallel processing,
When interference occurs, permission is given only to the earliest process, so that the overall processing speed increases and no deadlock occurs.

In the above embodiment, one processor executes one process. However, even when one processor executes a plurality of processes, the same applies if the processes are performed in a time-division manner. is there. However, while a process is accessing the lock management device, it is assumed that a task switch does not occur in the processor in which the process is being executed.

The present invention is not limited to the above embodiment, but can be implemented in other appropriate modes by making appropriate design changes.

[Effects of the Invention] Since the present invention is a resource lock management device for a computer system as described in the claims, even if a plurality of processors simultaneously access a resource lock management table in a multiprocessor environment, the deadlock occurs. There is no danger of falling into a lock, and reading and writing of the lock management table can be executed in parallel. Therefore, access to the lock management table can be prevented from becoming a system bottleneck,
System performance can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a resource lock management device of a computer system according to the present invention, FIG. 2 is a block diagram showing details of an access arbitration device, and FIG. 3 is a diagram showing access to a lock management table. FIG. 4 is an explanatory diagram showing a process state transition, and FIG. 4 is an explanatory diagram showing a configuration of a lock management table shared in a conventional example. 1a, 1b 1n Processor 7 Lock management device 8 Access arbitration device 9,10 Copy storage device 15 Monitoring device

Claims (1)

(57) [Claims] 1. A resource lock management device which has a plurality of resources shared by a plurality of processes executed by a plurality of processors, locks a specific resource to another process, and occupies the specific process. A plurality of lock management tables provided in a number smaller than the number of the plurality of processes; and means for selecting and allocating an unused lock management table among the plurality of lock management tables to a process access request. , Detect which resources each process is requesting,
Based on the detection result, means for permitting a lock process for a process that has issued an earliest access request and a process that has issued an access request that does not interfere with an access request of another process; Means for simultaneously updating a lock result by the permitting means with respect to another lock management table.