JP2795312B2 - Inter-process communication scheduling method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a time sharing OS (Ope) operating on a multiprocessor system.
(Rating System), which relates to an inter-process communication scheduling method for scheduling inter-process communication in the time sharing OS.

[0002]

2. Description of the Related Art In a multiprocessor system, various performances of a computer system are improved by using a plurality of processors to increase parallelism.

[0003] Problems that occur in a multiprocessor system include the following problems that occur in interprocess communication.

That is, communication data (data to be subjected to inter-process communication) is stored in a cache memory of a processor in which a transmission-side process (a process from which communication data is transmitted) operates, but is received in a reception-side process (data). Since the processor on which the communication data is transmitted is not always on the cache memory of the processor on which the communication data operates, the processor on which the receiving process operates must fetch data from the main memory, resulting in a large cache penalty. This is the problem.

[0005] An example of a solution to this problem is described in Japanese Patent Application Laid-Open No. 4-195576 (publication relating to "cache memory system"). In the multiprocessor system described in this publication, a cache memory including a decoder for judging inter-processor communication access and a dedicated memory for identifying inter-processor communication data for each cache block is provided. , Each processor has. Then, when the inter-processor communication (inter-process communication) is issued, the data (communication data) is temporarily stored in the cache memory, and then is automatically transferred to the transmission destination cache memory. Higher speed has been achieved. Here, the communication data on the cache memory is not overwritten unless the receiving process discards it.

That is, this multiprocessor system attempts to perform high-speed inter-process communication by transferring the contents (communication data) of the cache memory on the transmission side to the cache memory of the transmission destination.

[0007]

The inter-process communication in the above-described conventional multiprocessor system has the following problems.

A first problem is that the cache memory in which the communication data is written is occupied by the communication data until the communication data is discarded. Cannot use that cache memory "
This is a problem that performance (access performance that should be originally improved by using a cache memory) in the multiprocessor system is deteriorated.

A second problem is that every time the processor on which the process on the receiving side operates changes to another processor, it is necessary to perform inter-process communication for notifying the change. The problem is that the speed of inter-process communication becomes slow. The reason why such a problem arises is that the following functions in the time sharing OS of the multiprocessor system exist as prerequisites. When a process runs out of time slices allocated to one processor, the processor is relinquished to another process. Because of the “multiprocessor system”, when the receiving process that has used up the time slice is executed again, the receiving process does not always operate on the same processor as the previous process.

In view of the above, an object of the present invention is to control a transmitting process and a receiving process related to certain communication data to operate on the same processor as much as possible,
Inter-process communication scheduling that enables effective use of communication data in the cache memory to perform inter-process communication without the need for a special cache memory, thereby realizing high-speed inter-process communication in a multiprocessor system. In providing a method

[0011]

According to the present invention, there is provided an inter-process communication scheduling method, wherein a "priority", a "processor number", and a "transmission" are assigned to each process in a time-sharing OS operating on a multiprocessor system. A process management data storage area for storing process management data holding a "time"; and, when transmitting communication data, a transmission side to a "processor number" in process management data corresponding to a reception side process in the process management data storage area. A process that writes the number of the processor on which the process was running, writes the time at which the communication data was transmitted to the “transmission time” in the process management data, and issues an instruction to register the receiving process in the dispatch table Communication control means and the process management data when selecting a process. Scheduling by referring to the "processor number" in the process management data corresponding to each process in the storage area so that both the transmitting process and the receiving process in the inter-process communication operate on the same processor; and The transmission side process refers to the “transmission time” in the management data and performs the scheduling in consideration of the lapse of time from the transmission of the communication data to the present, and based on an instruction from the inter-process communication control unit, Scheduling control means for registering the receiving process in the dispatch table.

[0012]

Next, the present invention will be described in detail with reference to the drawings.

FIG. 7 is a block diagram showing an example of the configuration of a computer system (multiprocessor system) to which the inter-process communication scheduling method of the present invention is applied. This computer system represents a tightly-coupled multiprocessor system in which a main memory 3 is shared among all processors 1 and includes a processor 1, a cache memory 2, and a main memory 3. Note that the multiprocessor system to which the present invention is applied
It is not necessarily limited to a tightly coupled multiprocessor system.

(1) First embodiment

FIG. 1 is a block diagram showing the configuration of a first embodiment of the inter-process communication scheduling method according to the present invention.

The inter-process communication scheduling method according to the present embodiment is applied to a computer system as shown in FIG. 7, and includes a processor 1, a cache memory 2, a main memory 3, and a process processing unit 4. ing.

The main memory 3 includes a dispatch table 31 storing a plurality of executable processes existing in the computer system according to priorities (the concept of the priorities is the same as that in the prior art). A process management data storage area 32 for storing process management data existing for each process, and communication data 33 to be subjected to inter-process communication are held.

Each process management data in the process management data storage area 32 includes the “priority” of the process and the number of the processor 1 on which the transmitting process is operating in the inter-process communication in which the process is the receiving data. It includes a “processor number” indicating (identification information) and a “transmission time” indicating the time at which the communication data 33 in the inter-process communication was transmitted.

When the processor 1 accesses the main memory 2, it always goes through the cache memory 2.

The process processing section 4 includes an inter-process communication control means 41 and a scheduling control means 42 which operate under program control.

When transmitting the communication data 33 (when operating for the processor 1 on which the transmitting process is executed), the inter-process communication control means 41 operates the transmitting process for transmitting the communication data 33. The number of the processor 1 is written into the “processor number” in the process management data corresponding to the receiving process of the communication data 33, and the time at which the communication data 33 was transmitted is referred to as the “transmission time” in the process management data. And instructs the scheduling control means 42 to register the receiving process in the dispatch table 31.

When receiving the communication data 33 (when operating for the processor 1 on which the receiving process is executed), the inter-process communication control means 41 waits until the communication data 33 is transmitted. Put the receiving process on standby (until it receives it).

The scheduling control means 42 controls the processor 1 to select and retrieve a process to be executed next from the dispatch table 31 and to register a newly executable process in the dispatch table 31. It is performed in accordance with the priority of etc.

A feature of the first embodiment of the present invention is that when the scheduling control means 42 extracts a process from the dispatch table 31, that is, when the processor 1 selects a process to be executed next, it corresponds to each process. In other words, in addition to "priority" in each process management data, "processor number" and "transmission time" in each process management data are referred to.

FIG. 3 is a flowchart showing the processing of the inter-process communication control means 41 in this embodiment. This process
It comprises a data transmission determination step A1, a processor number / transmission time writing step A2, a receiving process registration instruction step A3, a data reception determination step A4, and a standby step A5.

FIG. 4 is a flowchart showing the processing of the scheduling control means 42 in this embodiment. This processing is performed in a dispatch table registration instruction presence / absence determination step B
1, a dispatch table registration step B2, a highest priority process selection step B3, a processor number determination step B4, a process execution step B5, a current time / transmission time comparison determination step B6, and a next candidate process selection step B7. Consists of

Next, the operation of the inter-process communication scheduling method according to the present embodiment configured as shown in FIG. 1 will be described in detail with reference to FIGS.

First, the operation of the inter-process communication control means 41 in the process processing section 4 will be described.

First, when a process operating in a certain processor 1 receives communication data 33 transmitted from another process (the determination in step A1 is “No (No)”).
And the determination in step A4 is “yes (yes)”. In the above description, the “process running on a certain processor 1” corresponds to the receiving process, and the “other process” corresponds to the transmitting process.

In this case, the inter-process communication control means 4
1 causes the processor 1 to wait until the communication data 33 is transmitted (until the communication data 33 is received) (step A5). At this time, the time-sharing OS to which the inter-process communication scheduling method of the present embodiment is applied yields the processor 1 to another process so that the other process is executed.

Next, consider a case where a process running on a certain processor 1 transmits communication data 33 to another process (when the determination in step A1 is "yes"). In the above description, “the process running on a certain processor 1” corresponds to the transmitting process, and “other process” corresponds to the receiving process.

In this case, the inter-process communication control means 4
1 first recognizes which process is the receiving process of the communication data 33.

The inter-process communication control means 41 stores the "processor number" and "processor number" in the process management data (predetermined process management data in the process management data storage area 32) corresponding to the receiving process recognized as described above. In the “transmission time”, the number of the processor 1 in which the transmitting process is operating and the current time are written (step A).
2).

Further, the inter-process communication control means 41
An instruction to register the waiting receiving process in the dispatch table 31 is issued to the scheduling control unit 42 so that the communication data 33 is received by the receiving process (step A3).

Second, the operation of the scheduling control means 42 in the processing section 4 will be described.

The case where the scheduling control means 42 is activated is as follows. In the above step A3, the inter-process communication control means 41
When an instruction is received from any processor When an instruction is given to pass a process to be executed next from any processor 1

When activated, the scheduling control means 42 determines whether the above case is met (whether or not an instruction to register in the dispatch table 31 has been received from the inter-process communication control means 41). Is determined (step B1).

If the scheduling control means 42 determines in step B1 that "an instruction to register in the dispatch table 31 has been received from the inter-process communication control means 41", the scheduling process of the receiving process according to the given instruction is performed. According to the priority, the waiting receiving side process is registered in the dispatch table 31 (step B2).

On the other hand, if it is determined in step B1 that "the instruction to register in the dispatch table 31 has not been received from the inter-process communication control means 41", the scheduling control means 42 performs a series of processing as described below. I do.

The time when one process is selected from the group of processes registered in the dispatch table 31 and the process is executed by the processor 1 (timing) is at the following times. When the process running on the processor 1 runs out of the allocated time slice, the process running on the processor 1
When the processor 1 enters a standby state due to waiting for O (Input / Output) or the like and surrenders the processor 1

In such a case, the processor 1 issues an instruction (request) to the scheduling control means 42 to pass the next process to be executed. Then, the scheduling control means 42 performs a series of processes as described below based on this instruction.

First, from the dispatch table 31, the first
The candidate process (the process with the highest priority) is selected (step B3).

Next, the "processor number" in the process management data (predetermined process management data in the process management data storage area 32) corresponding to the process selected in step B3 is the value of the processor 1 which has issued the request (instruction). It is determined whether it is a number (step B4).

If it is determined in step B4 that both numbers (both the "processor number" in the process management data and the number of the processor 1 which issued the request) are the same, the process (the current candidate) Process) is executed by the processor 1 (processor 1 that has issued the request) (step B5).

On the other hand, in step B4, "both numbers are different"
Is determined, this time, the “transmission time” in the process management data is compared with the current time to determine whether or not “current time ≦ transmission time + threshold” (step B6). . Here, the threshold value is a preset constant.

If it is determined in step B6 that "current time> transmission time + threshold", the process is executed by the processor 1 (step B5).

On the other hand, if it is determined in step B6 that “current time ≦ transmission time + threshold value”, a process having the second highest priority next to the current candidate process is selected from the dispatch table 31, and that process is selected. A new candidate process is set (step B7). Then, the process returns to step B4, and repeats the above-described processes from step B4.

Next, the effect of the first embodiment of the present invention will be described.

In this embodiment, as much as possible by checking the "processor number" in the process management data corresponding to the candidate process in step B4 in FIG.
The receiving-side process is operated on the processor 1 on which the transmitting-side process has been operating (the processor 1 with a high probability that the communication data 33 exists on the corresponding cache memory 2). As a result, the receiving process can access the communication data 33 which is likely to exist in the cache memory 2 corresponding to the processor 1 on which the receiving process operates (the receiving process can access the communication data 33 at high speed). And the speed of inter-process communication can be increased).

In the present embodiment, step B in FIG.
In step 6, the current time is compared with the “transmission time” in the process management data corresponding to the candidate process. Since the data in the cache memory 2 is sequentially rewritten with the passage of time, it is not advisable to stick to executing the transmitting process and the receiving process on the same processor 1 even after a long time has elapsed. . Therefore, in order to determine how much the effect of the present invention (hereinafter referred to as “cache effect”) by sharing the cache memory 2 between the transmitting process and the receiving process has decreased over time, The above comparisons have been made. That is, if the difference between the current time and the “transmission time” is equal to or smaller than a certain threshold value (constant), it is determined that the above-described caching effect exists, and the transmission process and the reception process are executed by the same processor 1. To be executed. vice versa,
Threshold (constant) that has a difference between the current time and the "send time"
If it is larger, it is determined that the above cache effect cannot be expected, and the receiving process is executed by any processor 1. Thus, the load can be prevented from being biased to a specific processor 1.

(2) Second embodiment

FIG. 2 is a block diagram showing the configuration of a second embodiment of the inter-process communication scheduling method according to the present invention.

The inter-process communication scheduling method according to the present embodiment is applied to a computer system as shown in FIG. 7, and includes a processor 1, a cache memory 2, a main memory 3, and a process processing unit 4. ing.

The main memory 3 includes a dispatch table 31 storing a plurality of executable processes existing in the computer system according to priority, and a process management data storage storing each process management data existing for each process. An area 32 and communication data 33 to be subjected to inter-process communication are held.

In the inter-process communication scheduling method of the present embodiment (second embodiment), unlike the first embodiment,
Each process management data in the process management data storage area 32 holds “running time” together with “priority”, “processor number”, and “transmission time”. Here, the “running time” indicates a time during which the transmitting process that has transmitted the communication data 33 to the process (receiving process) corresponding to the process management data has been operating in the processor 1.

The process processing section 4 includes an inter-process communication control means 41 and a scheduling control means 42 which operate under program control.

FIG. 5 is a flowchart showing the processing of the inter-process communication control means 41 in this embodiment. This process
It comprises a data transmission determination step A1, a processor number / transmission time writing step A2, a receiving process registration instruction step A3, a data reception determination step A4, a standby step A5, and a running time writing step C1.

FIG. 6 is a flowchart showing the processing of the scheduling control means 42 in this embodiment. This processing is performed in a dispatch table registration instruction presence / absence determination step B
1, a dispatch table registration step B2, a highest priority process selection step B3, a processor number determination step B4, a process execution step B5, a current time / transmission time comparison determination step B6, and a next candidate process selection step B7. And a threshold calculation step D1.

Next, the operation of the inter-process communication scheduling method according to the present embodiment configured as shown in FIG. 2 will be described in detail with reference to FIGS.

The operations in steps A1 to A5 in FIG. 5 and the operations in steps B1 to B7 in FIG. 6 are the same as those in the first embodiment. Therefore, description of the operations related to these processes is omitted.

The following is a description of an operation characteristic of this embodiment (an operation different from that of the first embodiment).

In the first embodiment, the threshold value used in the comparison judgment in step B6 is a preset constant. However, in the second embodiment, the threshold value is the time during which the transmitting process has been operating on the processor 1,
That is, the communication data 3 transmitted from the transmitting process
3 is calculated as a function of the "running time" in the process management data corresponding to the receiving process that receives "3." Therefore, the scheduling control means 42 calculates a threshold value from the “running time” (step D in FIG. 6).
1) A comparison is made in step B6 based on the threshold value. As the function, for example, when the value of the “travel time” is t, the constant is a, and the function (function indicating the threshold value) is f (t), f (t) = at
Can be considered.

In order to cause the scheduling control unit 42 to perform the process in step D 1, the inter-process communication control unit 41 writes the “processor number” and the “transmission time” in step A 2, Is executed in the "running time" in the process management data (predetermined process management data in the process management data storage area 32) corresponding to the receiving process (step C1 in FIG. 5).

The longer the transmission side process has been executed by the processor 1, the more the communication data 33 can be expected to exist in the cache memory 2 corresponding to the processor 1. Therefore, in the inter-process communication scheduling method of the present embodiment (the second embodiment), the threshold value is calculated as a function of the “running time” as described above, and the “judgment of the cache effect” is performed more accurately. I have. This allows
In the present embodiment, a unique effect that “the load balance tends to be more uniform among the plurality of processors 1” occurs.

[0065]

As described above, according to the inter-process communication scheduling method of the present invention, the following effects are produced.

The first effect is that by performing process scheduling such that both the transmitting process and the receiving process operate on the same processor (a processor corresponding to the same cache memory), the receiving process becomes An advantage is that communication data that is likely to remain on the cache memory can be accessed, and that communication data can be accessed at high speed, so that the speed of inter-process communication can be increased.

The second effect is that when the “transmission time” is set in the process management data and the current time is compared with the “transmission time” and it is determined that the cache effect cannot be expected, the receiving process can be executed by any processor. By scheduling a process so that it can be executed, an effect is obtained that a load is not concentrated on a specific processor. The second effect is, in particular,
Due to the fact that the “running time” in the process management data can be taken into account in the above comparison, the invention according to claim 2 is more remarkably generated.

That is, according to the present invention, during a period in which communication data in the cache memory is valid, the transmitting process and the receiving process operate on the same processor,
By making any processor executable after the valid period, the above two effects can be achieved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a first embodiment of an inter-process communication scheduling method according to the present invention.

FIG. 2 is a block diagram showing a configuration of a second embodiment of the inter-process communication scheduling method according to the present invention.

FIG. 3 is a flowchart showing a process of an inter-process communication control unit in FIG. 1;

FIG. 4 is a flowchart showing a process of a scheduling control unit in FIG. 1;

FIG. 5 is a flowchart showing the processing of the inter-process communication control means in FIG. 2;

FIG. 6 is a flowchart showing a process of a scheduling control unit in FIG. 2;

FIG. 7 is a block diagram showing a configuration of an example of a computer system to which the inter-process communication scheduling method shown in FIGS. 1 and 2 is applied;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Processor 2 Cache memory 3 Main memory 4 Process processing unit 31 Dispatch table 32 Process management data storage area 33 Communication data 41 Inter-process communication control means 42 Scheduling control means

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁶ Identification code FI G06F 15/163 G06F 15/16 320K 310V

Claims (4)

(57) [Claims] 1. In a time sharing OS operating on a multiprocessor system, each process has a "priority", a "processor number",
And a process management data storage area for storing process management data holding the “transmission time”, and a “processor number” in the process management data corresponding to the receiving process in the process management data storage area when transmitting the communication data. The number of the processor on which the transmitting process was operating is written in the, the time at which the communication data was transmitted is written in the "transmission time" in the process management data, and an instruction to register the receiving process in the dispatch table is issued. An inter-process communication control means to be issued; and a transmitting-side process and a receiving-side process in inter-process communication by referring to a “processor number” in process management data corresponding to each process in the process management data storage area when selecting a process. Are scheduled to run on the same processor. And the scheduling is performed in consideration of the lapse of time from the time when the transmitting process transmits the communication data with reference to the “transmission time” in the process management data, and the inter-process communication control is performed. Scheduling control means for registering a receiving process in a dispatch table based on an instruction from the means. 2. A process management data storage area for storing process management data which also holds a "run time" for each process, and a process corresponding to a receiving process in the process management data storage area when transmitting communication data. An inter-process communication control unit that writes the time at which the transmitting process was executed by the processor in the “running time” in the process management data, and “from the time when the transmitting process transmitted the communication data until the present time” when considering the process selection. And a scheduling control means for calculating a threshold value used in the determination of “elapse of time” based on “running time” in the process management data in the process management data storage area. The described inter-process communication scheduling method. 3. When f (t) indicating a threshold value calculated based on t of the value of the “travel time” is a constant, f
3. The inter-process communication scheduling method according to claim 2, wherein (t) = at. 4. The scheduling method according to claim 1, wherein the scheduling method is applied to a tightly coupled multiprocessor system.