JPH0754497B2 - Method for controlling a shared input/output device - Google Patents

Description

[Detailed Description of the Invention] (Industrial Application Field) The present invention relates to a control method for a shared input/output device, and more particularly to a control method for controlling the priority of input/output requests when multiple input/output requests from each client conflict in a system in which an input/output device is shared by multiple independent devices (hereinafter referred to as clients) such as microprocessors, which individually and at a given time output only one input/output request with the highest priority and do not output a new input/output request until the processing of the previous request is completed.

(Prior Art) Conventionally, in this type of shared input/output device system, when multiple input/output requests from clients conflict within a short period of time, a priority control method has been adopted to control the input/output requests according to the priority of these requests.

In FIG. 3, 100 is a shared input/output device, 20
Each of the clients 200, 300, and 400 is a microprocessor-like client that individually outputs only one I/O request at a time to the shared I/O device 100 and does not output a new I/O request until the processing corresponding to that request is completed. When making an I/O request to the shared I/O device 100, the clients 200, 300, and 400 add their priorities 21, 31, and 41 to the current request 201, 301, and 401, respectively, and output the I/O requests 210, 310, and 410.
is output.

The scheduler 101 in the shared I/O device 100 schedules the I/O requests 210, 310, and 410 based on the priorities 21, 31, and 41. If the priority 21 is the highest, followed by 31 and 41, the following order will be used: requests 201, 301, 410, and 410.
The input/output process 102 for the input/output medium 500 is executed according to the schedule in the order of 01.

(Problem to be Solved by the Invention) However, according to this control method, each client 200, 30
0, 400, respectively,
For example, if the priority of requests 201, 202, 301, 302, 401, 402, is 201 > 202 > 301 > ... (the left side has higher priority), the highest priority request 20
If there is a time lag between when the client 200 knows by some means that the processing for 1 has been completed and when it outputs the next request 202, another client
The processing for request 301 in 300 is prioritized,
In a different order than priority, i.e. request 201 → 301 → 202
There is a problem in that the processing is carried out in the order →...

For example, as shown in FIG. 5, at the time when the I/O processing corresponding to one I/O request in the shared I/O device 100 is completed, the current request 20 is
If the I/O process corresponding to 1 is completed, at this point, the shared I/O device 100 sends the next request 20 to the client 200.
2 occurs, and the next request 302 of another client 300 is not recognized as a current request 301, as shown in d.
and starts the corresponding process.

In the figure, 2A(10), 3A(11), and 4A(12) are written above the arrows indicating the current requests 201, 301, and 401 and the next requests 202 and 302.
The symbols (1), 4A (3), 2B (9), and 3B (5), as well as the 10, 9, 1, 5, and 3 written above the arrows indicating priorities 21, 31, and 41, have meanings as shown in the legend in the margin of the figure. For example, 2A (10) indicates the first request from client 200, with a priority value of 10. Similarly, 3B (5) indicates the second request from client 300, with a priority value of 5, and 4A (3).
indicates the first request from the client 400, which has a priority value of 3. Here, the higher the priority value, the higher the priority.

As a result of the above, as shown in the bottom row of Figure 5, the input/output processing time is 3B(5) → 2B between timings
In other words, the request 202 of the client 200, which should have been processed first in terms of priority, is executed in the order of (9).
(2B(9)) rather than the request 302 of the client 300 (3B
(5)) will be processed first.

In such a case, if an attempt is made to control the entire system to satisfy the desired priority by processing 601 (see FIG. 3) such as communication and exclusive control between the clients 200, 300, and 400, the logic becomes complicated, the independence between the clients is lost, and the processing associated with scheduling becomes complicated, resulting in an increase in overhead.

The present invention has been proposed to solve the above problems, and its object is to provide a method for controlling a shared I/O device that can properly control the priority order of multiple I/O requests to the shared I/O device throughout the entire system without losing the independence of each client and without increasing overhead.

(Means for solving the problem) In order to solve the above problem, the present invention provides a method for controlling a shared input/output device shared by a plurality of independent clients, each of which individually and at a given time outputs only one input/output request with the highest priority and does not output a new input/output request until processing of the highest priority request is completed, wherein, when each of the clients outputs the highest priority request of all currently occurring requests to the shared input/output device as a current request, the client adds the priority of the current request and the priority of the request for which output of the next input/output request is scheduled in the client to the current request and sets them in the shared input/output device, the shared input/output device refers to all the priorities added to and set on requests from each client, and executes input/output processing corresponding to the current request that has the highest priority among the priority of the current request and the priority of the request for which output is scheduled to be next, and after the input/output processing is completed, refers again to all the priorities added to and set on requests from each client, excluding the priority of the completed input/output request, and executes the processing corresponding to the current request that has the highest priority.
If the output of an I/O request is the priority of the scheduled request,
The I/O processing corresponding to the next I/O request is executed in the order of priority, by waiting for the output of the next I/O request from the corresponding client and then executing the I/O processing corresponding to that request.

When a need arises to change the priority order of the priority already set in the shared input/output device due to a change in a request scheduled to be next output from a client while the shared input/output device is executing input/output processing corresponding to a certain request, it is desirable to be able to reset the priority of the next request in the shared input/output device.

(Function) (1) In the present invention, each client passes to the shared input/output device its current request, the priority of this request, and, if a next request has already occurred, the priority of that request.

(2) Then, on the shared I/O device side, among the information (1) passed from all the clients, the "priority of the current request" is compared with the "priority of the next request" to determine the one with the highest priority. Regardless of whether the one with the highest priority is the "priority of the current request" or the "priority of the next request," the current request included in the I/O request from the client that output it is output to the I/O medium for processing.

At this stage, if "priority of next request">"priority of current request" (priority of next request is higher than priority of current request), the current request is pushed aside by "priority of next request" and input/output is performed. Also, when input/output is completed, if "priority of next request" is also higher than both "priority of current request" and "priority of next request" output by other clients, after input/output processing of the current request is completed, the I/O request for the next request from the client is output and it is input/output to the I/O medium.

(2-1) If only one request (i.e., only the current request) has occurred within a particular client, the "priority of the next request" is not passed to the shared I/O device.
In claim 1, the current request is compared with priorities from other clients by its own "current request priority", and the above-mentioned pushing operation and waiting for the output of the I/O request for the next request do not occur.

(2-2) In the case of an I/O request when multiple requests have already occurred within a particular client, the requests are arranged in order of priority within the client, so that "priority of current request">"priority of next request" (the priority of the current request is higher than the priority of the next request), and in accordance with claim 1, a push-out operation does not occur, but there may be a wait for the output of the I/O request for the next request.
However, in this waiting state, the I/O for the current request has already been completed, and the corresponding client will surely output the next request as a new I/O request, so the waiting state will not continue forever.

(2-3) If a new request with a higher priority than the "priority of the current request" occurs before an I/O request has already been output to the shared I/O device in the above (2-1) or (2-2) and the processing is completed on the shared I/O device side, the request is rearranged in the client's priority order, and the new request becomes the "next request." This is a method for simplifying the processing because the current request has already been output to the shared I/O device side, and it is generally unknown when the priority will reach its maximum and I/O processing will begin, and attempting to accurately recognize this would require exclusive control between the client and the shared I/O device, which would complicate processing and reduce performance.

The priority of the "next request" is passed to the shared input/output device by the reset process of claim 2, and when that priority becomes higher than the priorities of other clients, the above-mentioned push-out operation occurs.

When a push operation occurs and the I/O processing corresponding to that I/O request is completed, the client that generated the I/O request marks the previous "next request" as the "current request."
However, the fact that a client is pushed out indicates that the "next request priority" of that client, i.e., the new "current request priority" is the highest priority among all the clients, so the shared I/O device waits for that I/O request and performs I/O on the I/O medium. Also, this waiting state does not become a permanent waiting state because the I/O of the current request has already been completed and the corresponding client will always output the next request as a new I/O request.

(Embodiments) Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 shows the configuration of this embodiment.
Reference numeral 200, 300, and 400 denote a shared I/O device, 200, 300, and 400 denote clients such as microprocessors, and 500 denotes an I/O medium of the shared I/O device 100. In each of the clients 200, 300, and 400, a current request 201, 301, or 401 occurs, and thereafter, requests 202, 303, and 404 occur.
Assume that occurrence of requests 201, 301, 401, 202, 302, and 402 is scheduled, respectively, and these requests are output to the shared I/O device 100 as I/O requests 210, 310, and 410. Here, the priority of the requests 201, 301, 401, 202, 302, and 402 is as follows: request 201>202
> 401 > 402 > 301 > 302 (the ones on the left have higher priority).

Thus, when a client 200 makes a current request 201 to the shared I/O device 100, it adds a priority 21 to the current request 201 and also schedules the next request 201 within the client 200.
If request 02 has already occurred, then a priority of 22 is also added to request 202, which has the next highest priority after request 201.
This process is similar for the other clients 300, 400, and priorities 31, 32 and 41, 42 are added when the requests 301, 401 are output from the respective clients 300, 400.

In this case, the shared I/O device 100 first processes the requests 201, 301, 401, 202, 302, and 402 with priorities 21, 31, 41,
22, 32, and 42 are compared, and the current request 201 having the highest priority is selected and executed by the scheduler 101. That is, the I/O device 102 for the I/O medium 500 is executed according to the request 201.

Next, when the execution of the request 201 is completed and rescheduling is to be performed, first the priorities 31, 41, 22, 32, and 42 of the remaining requests 301, 401, 202, 302, and 402 are compared. Then, if the highest priority among them is the priority 22 of the request 202 that is scheduled to output an I/O request and that was added to the completed request 201, the corresponding client 2
Wait for the next request 202 to be output from 00, and then
Execute the input/output process corresponding to 02.

In addition, when only one request (i.e., the current request 201) has occurred in a particular client, for example 200, the "priority of the next request" is not sent to the shared I/O device 100. Therefore, the shared I/O device 100 does not receive the current request 201.
01 will be compared with the priorities of other clients with its own priority of 21, and the above-mentioned pushing operation or waiting for the output of the I/O request for the next request will not occur.

In the embodiment, in the case of an I/O request when multiple requests 201 and 22 have already occurred in a particular client, for example 200, the requests are arranged in order of priority in the client 200, so the priority 21 of the current request 201 is higher than the priority 22 of the next request 202, and the above-mentioned pushing operation does not occur. However, when the processing corresponding to the current request 201 is completed,
Waiting for the output of an I/O request for the next request 202 may occur. However, in this waiting state, the I/O for the current request 201 has already been completed, and the client 200 is not necessarily waiting for the next request 202.
is output as a new I/O request, so the waiting state does not last forever.

Similarly, each time a request is executed, the remaining priorities in the shared input/output device 100 are compared, attention is paid to the request with the highest priority, and the above-mentioned process is repeated while waiting for that request to be output.

As a result, even if there is a time lag between the execution of a request with the highest priority and the output of another request with the next highest priority, there is no inconvenience of a request with a lower priority being executed first, and proper priority control can be performed through all of the clients 200, 300, 400. To explain the processing in this embodiment described above with reference to Fig. 4, as shown as a at time point a in Fig. 4, the shared input/output device 100 can recognize that the next request 202 from the client 200 has a priority value of 9, and can predict that this request 202, 2B(9), will be output as the current request 201 of the client 200, and wait for the output of that request before processing the next request 3 of the client 300 at b.
This allows the I/O processing for the request 202 to be performed prior to the I/O processing for 3B(5) which is 02. This allows the I/O processing for 2B(5) to be performed according to the priority value during the I/O processing time between .
The processing will be executed in the order of (9) → 3B(5).
The legend shown in FIG. 5 is also used in FIG.

Next, in FIG. 1, the priority order of requests 201, 202, 301, 302, 401, 402 of clients 200, 300, 400 to the shared input device 100 is 201>301>401>202>302>402 (priority 21>31>41>22>32>42), and each client
200, 300, and 400 make requests as in the previous example, and while the shared I/O device 100 is processing the request 201,
A new request with priority 23 is scheduled in 200, and the priority of all remaining requests becomes 203 > 301 > 401 > 202 > 302 > 402 (priority
The processing when the order changes to 23>31>41>22>32>42 is explained below.

In this case, as shown in FIG. 2, the client 200:
A priority reset 220 is performed for the request 203, which has the second highest priority after the request 201, for the shared input/output device 100. That is,
Priority 22 for the previous next highest requirement 202 has been changed to requirement 203.
The priority level for this item will be changed to 23.

That is, by rearranging the requests in the client 200 in order of priority, the request 203 becomes the "next request" and
The push-out condition occurs when the "next request priority" reset process causes the priority 23 for request 203 to be higher than the priority of other client requests (eg, priority 31 for request 301).

When a push-out state occurs and the I/O processing corresponding to the I/O request 201 is completed, the client 200 that generated the I/O request 201 outputs a new I/O request 203, with the previous "next request" as the "current request". However, the occurrence of the push-out itself is not recognized by the client.
Since the "next request priority" of 200, i.e., the new "current request priority" is the highest priority among the priorities output by all clients, the shared I/O device 100 waits for the I/O request 203 to be output and then performs I/O on the I/O medium 500. Also, even while waiting for the output of this I/O request 203,
Since the I/O of the current request 201 has already been completed, and the corresponding client 200 will always output the next request 203 as a new I/O request, there will be no perpetual waiting state. This makes it possible to perform proper scheduling based on the priority of each request, regardless of changes in the circumstances of each client.

Incidentally, the present invention is of course applicable to a multiprocessor system having a common bus, as well as to a system in which a plurality of independent microprocessors are connected via a communication line.

(Effects of the Invention) As described above, according to the present invention, the shared input/output device performs scheduling by referring to the priorities of all current requests and subsequent requests generated by each client, and also makes it possible to reset priorities as necessary, so that proper priority control across the entire system can be simply realized without requiring communication between clients or exclusive control as in the conventional method.

Therefore, there is no need to worry about losing the independence of each client, and there are other advantages such as the ability to reduce the overhead of shared input/output control.

[Brief description of the drawings]

Fig. 1 is a diagram of an input/output device sharing system to which an embodiment of the present invention is applied, Fig. 2 is an explanatory diagram of a case where a new request occurs in a client, Fig. 3 is a diagram of an input/output device sharing system showing a conventional example, Fig. 4 is an explanatory diagram of processing in the embodiment of the present invention, and Fig. 5 is an explanatory diagram of processing in the conventional example. 100... Shared input/output device 200, 300, 400... Clients 201, 202, 203, 301, 302, 401, 402... Requests 210, 310, 410... Input/output requests 21, 22, 23, 31, 32, 41, 42... Priority

Claims (2)

[Claims] [Claim 1] A method for controlling a shared input/output device shared by a plurality of independent devices (hereinafter referred to as clients), each of which individually and at a given time outputs only one input/output request with the highest priority and does not output a new input/output request until processing of the highest priority request is completed, wherein, when each of the clients outputs the highest priority request of all currently occurring requests to the shared input/output device as a current request, the client adds the priority of that request and the priority of a request for which the client is scheduled to next output an I/O request to the current request and sets them in the shared input/output device, the shared input/output device refers to all the priorities added to and set on requests from each client, and executes I/O processing corresponding to the current request that has the highest priority among the priority of the current request and the priority of the request for which the next output is scheduled, and after the I/O processing is completed, the shared input/output device again refers to all the priorities added to and set on requests from each client excluding the priority of the completed I/O request, and if the highest priority is the priority of a request for which the output of an I/O request was scheduled that was added to the completed I/O request, waits for the output of the next I/O request from the corresponding client and executes the I/O processing corresponding to that request, thereby executing I/O processing corresponding to the requests in order from highest priority. " 2. While a shared input/output device is executing an input/output process corresponding to a certain request, a change in a request scheduled for output of the next input/output request that occurs in a certain client causes:
When it becomes necessary to change the priority order of the priority already set in the shared input/output device, the priority order of the next request can be reset in the shared input/output device.
A method for controlling the shared input/output device according to claim 1.