JPH04330529A - Management system for shared resource - Google Patents

Abstract

PURPOSE:To effectively use a shared resource without repeating ineffective execution. CONSTITUTION:Plural tasks, processings, or processors PEi request the use of a shared resource 3 on a network 1 to output a packet 11 having a specific tag 21, and the shared resource 3 in the available state outputs a packet 10 having a tag 21 designating itself, and plural arrival packets 11 are stored in a queue memory 4 on the network 1, and one of plural packets 11 having the tag 21 matching with the tag 21 of the shared resource 3 is selected to output a permission packet which permits the use of the shared resource.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a shared resource management method for data-driven microprocessors.

0002

[Background Art] With the advancement of VLSI technology, the price of microprocessors has decreased and their reliability has improved, so parallel processing in which multiple microprocessors are connected to a network to process data at high speed has become popular. ing. However, if the processes divided among a large number of microprocessors are processed completely independently, problems may occur. In particular, when multiple microprocessors use shared resources such as storage devices or printers at the same time, the consistency of processing may not be guaranteed unless the microprocessors that can be used at the same time are restricted to one and executed exclusively. be.

Conventionally, when multiple microprocessors (tasks) use a shared resource exclusively, a series of access requests to memory areas indicating the availability of the shared resource are interlocked (concatenated). During the test, a test-and-set instruction was used to exclusively rewrite the unused indicator.

0004

[Problems to be Solved by the Invention] As shown in FIG. 6, conventionally, in step S1, a TEST & SET command is executed;
If step S2 is successful, shared resources are used in step S3. When the shared resource is being used by another task, the requesting task must return to step S1 and repeat invalid execution until the shared resource becomes unused and is successful. As a result, there was a problem in that the pipeline processing of the network system as a whole was delayed.

An object of the present invention is to improve the processing speed of the entire network system by making it possible to use shared resources without repeating invalid execution.

[0006]

[Means for Solving the Problems] In order to achieve the above object, in the present invention, a plurality of tasks, processes, or processors PEi request the use of a shared resource 3 on a network 1 and issue a specific tag 21. The shared resource 3 in an available state outputs a packet 10 having a tag 21 specifying itself, and the waiting memory 4 on the network 1 stores the plurality of arrived packets 11. Select one of the plurality of packets 11 having a tag 21 that matches the tag 21 of the shared resource 3,
A permission packet that permits the use of the shared resource 3 is output.

[0007]

[Operation] Tag 21 of packet 10 from shared resource 3,
The tags 21 of the packets 11 from the processor PEi etc. are compared in the waiting memory 4, and if there is a match, one is selected from the plurality of matching packets 11 to generate a permission packet. This permission packet is returned to the processor PEi that outputs the selected packet 11, so that this processor PEi can use the shared resource 3.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be explained below with reference to the drawings. In FIG. 1, 1 is a network, 2 is a management processor that manages shared resources, 3 is a shared resource such as a storage device or a printer, 4 is a waiting memory (MM), and 5 is an instruction execution unit. Further, PE1, PE2...PEn are processors connected to the network 1.

[0009] Shared resource 3 and each processor PE1, PE
2...PEn is the packet 10,1 as shown in Figures 2 and 3
1 to network 1. A number q (constant) is set in advance for the management processor 2, and sequential numbers E1, E2, . . . En are simply set for each processor PEi, etc. On the other hand, shared resource 3 also has a number k in advance.
(constant) is set and processor P on network 1
E1, PE2...PEn, the management processor 2, and the shared resource 3 each have the corresponding numbers q, k, E1, E2...E
It can be identified by n.

FIG. 2 shows a packet 10 that the shared resource 3 sends over the network 1 to the management processor 2 when it becomes available. First, to explain each bit section of the packet 10, 20 consisting of multiple bits from the high-order bit side indicates the number of the destination processor of this packet 10, and the number #q indicates that the management processor 2 is the destination processor. There is.

[0011] Next, 21 indicates a destination node number which is also used to identify a pair creation (mate) partner, and in particular, the shared resource 3
A predetermined value #k is set for the purpose. 22 is on the left,
An identification flag indicating right data, for example, 1 and 0 indicate left and right, respectively. In this case, since a packet containing right data is to be output from the shared resource side, a flag indicating right is set. Reference numeral 23 indicates an instruction code (operation code), which is, for example, an operation instruction such as addition or subtraction, and RET-REQ is set here. 24 is a bit area where other flags are set. A data area 25 includes information necessary for managing the shared resource 3 such as time and job number.

FIG. 3 shows a packet 11 indicating a request for the processor PEi to use the shared resource 3. This request packet 11 is not sent directly to the shared resource 3, but is directed to the management processor 2 (#q) that specializes in management. Therefore, the destination processor number 20 is the number q of the management processor 2, and the next destination node number 21 is #k, which is predetermined for the shared resource 3.

The left and right data identification flag 22 is set by the processor P.
A packet containing left data is to be output from the Ei side, so specify left, and instruction code 23 is R.
ET-REQ, and data including the identification number #Ei of the requesting processor PEi and the return destination node number are stored in the data area 25.

When the shared resource 3 becomes available, the packet 10 containing the information shown in FIG.
)4.

On the other hand, any process PEi that wants to use shared resource 3 sends packet 1 containing the information shown in FIG.
1 to management processor 2 (#q) via network 1
The data is sent to the waiting memory (MM) 4.

In the waiting memory 4 of the management processor 2, when the packets 10 and 11 of FIG. 2 and FIG. packet 10,
11 is sent to the instruction execution unit 5. The instruction execution unit 5 executes the instructions RET-REQ of packets 10 and 11, that is, the requesting processor PEi stored in the left data
Let the identification number Ei of Ei be the new destination processor number,
Furthermore, the return destination node number outputs a packet indicating permission to use the shared resource with the new destination node number. The identification number #Ei of the destination processor PEi and the return destination node number are data sent from the request source PEi, and the node number indicates the beginning of the process that uses the shared resource 3. As a result, the requesting PEi starts the process of using the shared resource 3. When the processing is completed, the shared resource 3 also sends the packet 10 of FIG. 2 to the management processor 2.

FIG. 4 shows a data flow graph of the above operation, in which a packet indicating that the shared resource can be used is output from the shared resource 3 as the right input, and a packet indicating that the shared resource can be used is input as the left input to the processor PEi.
A packet requesting the use of a shared resource is input and fired, and the round node issues a RET-REQ command and outputs a permission packet. The requesting processor PEi exclusively uses the shared resource 3 and performs usage processing. When multiple processes request the use of the shared resource 3, the MM4 of the management processor 2 recognizes the packets 11 with different request source processor numbers and return destination node numbers in the data 25 as packets 10 from the shared resource 3. We will have to wait for it to arrive. It is necessary to select one packet 10 in FIG. 2 from among the plurality of packets 11 in FIG. 3, and control the remaining packets 11 to continue waiting in the MM4. Next, one process performed by the MM 4 when a plurality of packets 11 that attempt to use the shared resource 3 from a plurality of processors PE1, PEi, .

FIG. 5 shows the detailed configuration of the waiting memory (MM) 4, in which 7 is a hash memory (HASH), 8 is a control section, 9 is a content addressable memory (CAM), 12 is a priority word selection section, and 14 is a timer. be. Multiple request packets 11
When the packet arrives at the MM 4, if there is a corresponding hash address, and if a packet is already stored at the corresponding hash address in the hash memory 7, the control unit 8 sequentially stores the packet in the associative memory 12.

Therefore, when the packet 10 indicating availability from the shared resource 3 arrives at the management processor 2 (q) via the network 1, the control unit 8 stores the hash memory 7
If it is checked whether packets 11 are stored in the corresponding hash address of
, 11 are generated as a pair. When the hash memory 7 is also waiting for other packets (other than resource usage), when the corresponding address is in an empty state, or when a packet with a different identifier and different from that of the mate is stored, the contents of the associative memory 9 are It is checked whether packet 11 to which packet 10 should be mate is stored. If a plurality of packets to be mate exist in the associative memory 9, the packet stored at the lowest address is selected by the priority word selection section and mate processing is performed. Another processing method performed by the waiting memory (MM) 4 when a request to use the shared resource 3 is issued from a plurality of processors PE1, PEi will be described.

When a plurality of request packets 11 arrive at the management processor 2, the control unit 8 uses a timer 14 to time their arrival times. When storing each packet 11 in the hash memory 7 or the associative memory 9, the control unit 8 records the packets 11 with the arrival times attached (time stamped) in each entry of the HASH 7 or CAM 9. In this case, a memory area for extra timestamps is reserved in advance for each entry. When the packet 10 indicating availability arrives at the MM 4 from the shared resource 3, the control unit 8 generates a pair from among the request packets 11 stored in the HASH 7 and the CAM 9, starting with the one with the oldest time stamp.

Regarding the use of the shared resource 3, yet another processing method is to provide a priority flag in the data 25 area of each packet 11 when each processor PE has a priority, or when there is a priority depending on the processing content. You can also do it. HASH7 and CAM9
The control unit 8 first checks the priority flags of the packets 11 waiting in the queue, and fires the packets 11 with a high priority as the available packets 10.

[0022] In the above embodiment, the management processor 2 (
q), but the shared resource 3 itself may have the function of a management processor. Further, in the above embodiment, the shared resources are provided outside the processor PE, but within one processor PE, the shared resources are provided outside the processor PE.
It may also be a memory in E.

Furthermore, in the above embodiment, each processor P
One request packet is issued from E1, PE2, etc., or each processor PE1, etc. may output a plurality of request packets 11. Furthermore, there are two or more shared resources 3 with the same function, and resource number 1 is included in the data 25 of packet 10 in FIG.
, 2, etc. may be stored and identified. Further, although the above embodiment shows an example in which shared resources are arranged on a network, the present invention is not limited to this, and can also be applied to the management of multiple tasks being executed by a single processor. Needless to say.

[0024]

As described above, according to the present invention, multiple tasks, processes, or processors request the use of shared resources on a network, output packets with specific tags, and use the shared resources. A shared resource in the enabled state outputs a packet with a tag that specifies itself, and a waiting memory on the network stores the multiple arriving packets and outputs a packet with a tag that matches the tag of the shared resource. Since one of them is selected and a permission packet for permitting use of the shared resource is output, the waiting memory automatically determines whether or not the shared resource can be used. Therefore, there is no need for each processor to repeatedly perform invalid execution like a test-and-set instruction, and other processes can be effectively executed during that time, greatly improving the processing capacity of the entire system.

[Brief explanation of the drawing]

FIG. 1 is an overall configuration diagram illustrating the present invention.

FIG. 2 is a diagram showing a packet configuration of the present invention.

FIG. 3 is a diagram showing a packet configuration of the present invention.

FIG. 4 is a flow graph showing the operation of the present invention.

FIG. 5 is a diagram showing the configuration of a waiting memory according to the present invention.

FIG. 6 is a flowchart showing conventional operation.

[Explanation of symbols]

1 Network 2 Management processor 3 Shared resource 4 Waiting memory 5 Instruction execution unit PE1, PE2...PEn Processor 10, 1
1 packet

Claims (4)

[Claims] 1. A plurality of tasks, processes, or processors requesting the use of a shared resource in a system and outputting a packet having a specific tag, the packet having a tag specifying the shared resource in a usable state. output the packet,
A waiting memory in the system stores the plurality of arrived packets, selects one of the plurality of packets having a tag that matches the tag of the packet corresponding to the shared resource, and selects one of the plurality of packets having a tag that matches the tag of the packet corresponding to the shared resource. A shared resource management method that outputs a permission packet that permits use. 2. The shared resource management method according to claim 1, wherein the selected one packet is a certain packet stored at a low address in the waiting memory. 3. The shared resource management method according to claim 1, wherein the selected one packet is a packet that arrived at the waiting memory early. 4. The shared resource management method according to claim 1, wherein priorities are determined in advance among the plurality of tasks, processes, or processors, and the selected packet is a packet with a high priority. .