JPH0215354A - I/o scheduling method - Google Patents

Abstract

PURPOSE:To distribute the load of a control device and to improve the through- put of I/O processing by generating an I/O request to a control device having minimum load out of control devices to be connected to an external storage device. CONSTITUTION:At the time of receiving an I/O request indicating a data name and a processing method from an I/O request part 2, an allocation deciding part 31 checks an external storage device number concerned from the I/O request. The deciding part 31 allocates the control device having minimum current load out of the control devices to be connected to the external storage device to the I/O request by referring a connecting information table 33 indicating which control device is connected to the external storage device concerned and current load information tables 34 for respective control devices and generates an I/O signal. On the other hand, an end processing means 32 informs I/O end to the originating source of the I/O signal at the time of generating an I/O end interruption and updates the load information of the table 34 corresponding to the control device ending I/O operation. Consequently, the load levels of the control devices are uniformed and the through-put of I/O operation can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an I/O scheduling method, and in particular to an I/O scheduling method.
The present invention relates to an I/O scheduling method in which allocation of I/O requests to control devices is determined in order to balance the load status of a group of control devices at the stage of issuing an I/O request.

[Conventional technology]

In conventional devices, when an I/O request is received, the channel, the data path between the channel and the control device, and the control device sequentially examine unused paths and issue an I/O request to the unused path. Ta. However, no consideration was given to scheduling methods to prevent imbalances in the load on the control device.

[Problem to be solved by the invention]

The above conventional technology checks whether the channel, the data path between the channel and the control device, and the control device are in use, searches for an unused path, and issues Ilo. In a control device that can handle multiple calculation requests, if the load is concentrated on a specific control device, there is a problem with the processing speed of I/O processing.

An object of the present invention is to provide an I/O scheduling method that distributes the load on a control device and improves the throughput of I/O processing.

[Means to solve the problem]

In order to achieve the above-mentioned object, the present invention sends I/O requests to each control device using information on the load status of the control device group and connection relationship information such as which control device each external storage device is connected to. The load information includes three types of information: the execution multiplicity of I/O requests, the expected workload indicating how many blocks of data will be transferred, and the expected required time calculated based on the actual processing speed of the control device and the amount of data transferred. The feature is that it is scheduled using .

[Effect]

In the present invention, the schedule section includes an allocation judgment section and a termination processing section, and when the allocation judgment section receives an I/O request indicating the data name and processing method from the r/○ request section, the
Check the applicable external storage device number from the O request, refer to the connection information indicating which control device the external storage device is connected to, and the current load information of each control device, and connect to the external storage device. Among the available control devices, the control device with the smallest current load is assigned to the I/O request, and a mark/○ is issued. Thereafter, the load amount due to the I/O request is estimated and the load information corresponding to the control device is updated. On the other hand, when an I/O completion interrupt occurs, the completion processing unit reports completion to the issuer of the I/O request, and updates the load information in the load information table corresponding to the control device for which ■/○ has been completed. Update.

With this method, I/O can be issued to a control device with a low load, so the load among the control devices can be balanced and the throughput of Ilo can be increased.

〔Example〕

An embodiment of the present invention will be described below.

Figure 1 is.

The functional configuration of this embodiment is shown. 2 is an engineering/○ request unit that starts the schedule unit 3; 3 is a schedule unit; a connection information table 33 indicating which control device a certain external storage device is connected to; and a load information table 34 for a certain control device. ．． An allocation determining unit 31 that allocates I/O requests to control devices based on the connection information table 33 and load information table 34 and issues I/Os, and a termination process that performs termination processing and a report processing unit to the issuer of work/○ It consists of part 32. FIG. 2 shows the hardware configuration. Reference numeral 1 is a CPU (central processing unit), 4 is a channel, and 5 is a control device that controls an external storage device 6 and performs data transfer and calculation. 6 is an external storage device in which data is stored. In addition, the control device [5 is
It has a request reception buffer 51 that receives and receives a plurality of calculation requests.

3 and 4 respectively show the connection information table 33.
and the contents of the load information table 34. FIG. 5 shows the contents of the I/O request block. 6th
8 to 8 are flowcharts showing the operation of the scheduler 3.

The operation of the scheduler 3 will be described below with reference to flowcharts shown in FIGS. 1 to 5 and 6 to 8.

When an I/O request is issued from the I/O request section 2 on the CPUI side, the schedule section 3 activates the allocation judgment section 31. (Steps 201 and 202 in FIG. 6) When the allocation determination unit 31 is activated, it first obtains the external storage device number 23 from the I/O request block 20. Next, the connection information table 33 is referred to, the load information table 34 is checked for all the control devices connected to the external storage device, and the control device with the smallest load (control device number specified) is checked. Steps 204, 205 and 206 in Figure 7)
. In this case, among the loads in the load information table 34, r/
O Requirement Determined by degree of demand, expected amount of work, and expected required time. Next, I
I/O issue time 27 is registered.

After that, 1 is added to the I/O request degree of the load information table 34.
(ah4-ah+1.Here, A4-B is assumed to substitute B for A), check the number of processing blocks 26 (assumed to be b) from the I/O request block 20, and calculate the expected work in the load information table 34. In addition to the amount (bh←b h + b
) This is also registered in the total number of blocks 28 waiting for processing in the I/O request block 20, and the processing time is calculated by dividing the number of processing blocks 26 by the actual processing speed (referred to as dk) in the load information table 34 and calculating the load. Estimated time required for information table 34 (
ck) in addition to (c k4-c b+ b x/
d k), updating the load information of the control device; lastly,
Point the I/O request block 20 to the queue managed by the load information table 34 (step 2 in FIG.
07°208.209).

Next, processing at the time of an I/O end interrupt will be explained.

Upon receiving the I/O completion report from the control device 5.

The CPUI issues an I/O end interrupt to the scheduler 3. When the scheduler 3 receives an I/O end interrupt, it activates the end processor 32.

(Steps 201 to 203 in FIG. 6) Termination processing unit 32
When started, first, it checks the point to the ■/○ request block in the load information table 34 from the processing request number reported by the I/O end interrupt and the control device number (assuming the control device number is j). A /O request block 20 corresponding to the processing request number is searched for and removed from the queue.

Next, the control device number j that points to the queue
Subtract 1 from the I/O request level of (aJ←aJ-1),
Subtract the number of processing blocks 26 for the I/O request block 20 (the number of processing blocks is assumed to be ) from the expected workload (bJ
4-bJ by), redefine the actual processing speed dJ and update the load information. The actual processing speed is defined as the I/O issue time 27 (ts) of the I/O request block 20, and the total number of blocks waiting for processing, 28 Cbz, where the time when the I/O end γ interrupt is received is t6. ), calculate as follows.

First, the number of blocks processed from the start to the end of Ilo is calculated as follows.

bz/(te-ts) The average of this and the previous real processing speed is set as the new real processing speed.

dJ=(da+bz/(te-ts))/2 Finally, a completion report is sent to the issuer of the ■/○ request. (2 in Figure 8
/O, 211, 212, 213) By the method described above, it is possible to balance the I/O request level, expected workload, and expected required time between control devices.

〔Effect of the invention〕

If the load is concentrated on a control device that can perform calculations or accept multiple requests during data transfer,
Although the flow rate of I/O processing is reduced, according to the present invention,
By providing load information for control devices, updating the load information at the time of I/O issue and I/O completion, and allocating I/O requests based on the load information, it is possible to distribute the load between control devices. Since the deviation can be reduced, it is effective in speeding up ■/○ processing.

[Brief explanation of the drawing]

Fig. 1 is a functional block diagram of an embodiment of the present invention, Fig. 2 is a hardware connection block diagram of the present invention, Fig. 3 is a diagram showing the contents of the connection information table, and Fig. 4 is the processing of the load information processing section. It is a figure showing a flow. 1...Host computer, 2...Engineering/○ request section, 3...
-Schedule section, 31... Allocation judgment section, 32...
Termination processing unit, 33... Connection information table, 34...
Load information table, 4...channel, 5...control device, 51...required diagram! ／ Character table Peff Y Izuzu Kayatozu grope♂ Zuzuzuzuzu etc. 7 pictures

Claims (1)

[Claims] 1. In a computer system having a plurality of control devices and a plurality of external storage devices that can perform a calculation during data transfer or accept a plurality of calculation conditions for the calculation, the load status of a group of control devices 1. An I/O scheduling method comprising the steps of: monitoring an external storage device; and issuing an I/O request to a control device with the least load among control devices that can be connected to an external storage device. 2. The I/O scheduling method according to claim 1, wherein the load status of each control device is the multiplicity of execution of I/O requests to the control device. 3. The I/O scheduling method according to claim 1, wherein the expected workload of each control device is used as the load status of each control device. 4. The I/O schedule according to claim 1, wherein the expected required time converted from the processing speed and data transfer amount of each control device is used as the load status of each control device. Uehling method.