JP2001147886A - Device and method for sharing disk time - Google Patents

Abstract

PROBLEM TO BE SOLVED: To automatically adjust the operation condition of time sharing fulfilling requested performance based on an achievement value. SOLUTION: An input and output schedule mechanism 20 defines the rate of a time for allowing each input and output group obtained by grouping the sources of input and output to a disk device 16 to use a disk, and decides an assignment time (quantum) for allowing each input and output group to continuously use the disk device based on the defined time rate, and operates time sharing for allowing the competing input and output groups to use the disk device by sequentially switching the assignment time at the time of receiving a request for the input and output from a plurality of input and output groups to the disk device. A tuning part 52 automatically adjusts the operation condition of the time sharing according to the requested performance and the achievements.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a disk time sharing apparatus and method for scheduling the use of a disk drive based on a plurality of inputs and outputs.
The present invention relates to a disk time sharing apparatus and a method for scheduling the use of a disk apparatus so as to sequentially change an allocation time for competing input / output.

[0002]

2. Description of the Related Art Conventionally, in a storage system that manages data using a disk device such as a hard disk drive, for example, a disk device is a device having a RAID configuration, and this RAID device is connected under a disk control device. To process input / output from a host at a higher level, or directly connect a RAID device to a server to process input / output from a server OS.

In such a storage system, it is necessary to perform random access for which the response time is required to be guaranteed and sequential access in which the processing amount per unit time is emphasized for the same disk device. In this case, operation is performed in different time zones so that random access and sequential access do not conflict with each other. For example, in the daytime, random access-oriented OLTP operations (On Line Transact
ion processing) and back up the database at night after the work is over.

[0004]

[Problems to be Solved by the Invention] However, in such a storage system, it is necessary to continue the random access OLTP business even at night due to the non-stop operation of the storage system. It is necessary to execute a backup that is a sequential access during the OLTP business.

In the case of only random access, IOPS (Input Output Per Second), which is the number of input / output operations per unit time, which can satisfy a certain average response time, for example, 30 ms,
For example, 100 IOPS can be estimated. In the case of only sequential access, a throughput of, for example, 20 MB / s can be estimated.

However, when random access and sequential access are performed simultaneously, the received I / O is processed in a queue using a FIFO, so that the time required for the random access to use the disk device and the time required for the sequential access to the disk device are reduced. There is no mechanism to guarantee usable time.

For example, 50 IOPS with an average response time of 30 ms
Even if random access and 5 MB / s sequential access are desired, if sequential access frequently occurs, the sequential access throughput does not need to be increased, but from 5 MB / s to 10 MB / s.
It goes up to MB / s. Conversely, the IOPS satisfying the average response time of 30 ms by random access decreases from 50 IOPS to 25 IOPS, although it is not desired to reduce it. 2. Resource Allocation Between Logical Volumes In a conventional storage system, data having different performance requirements is allocated to different disk devices to derive their respective performance characteristics. For example, data for which a response time is required to be guaranteed by random access of a small amount of data and data in which a processing amount per unit time is emphasized in a sequential access of a large amount of data are arranged in different disk devices.

However, with the increase in the capacity of disk devices, the number of cases where data having different performance requirements are arranged on the same disk device has been increasing. Similar problems occur when logical volumes having different performance requirements are arranged on the same disk. Conventionally, input / output
There is no mechanism for controlling disk resource distribution between logical volumes by scheduling in O. Therefore, if input / output to a certain logical volume frequently occurs, input / output performance to another logical volume is reduced.

For example, if a volume A for which 10 IOPS is to be guaranteed and a volume B for which 50 IOPS are to be guaranteed are arranged on the same disk, if access to volume A occurs frequently, the IOPS of volume A need not be increased. Despite none, it goes up from 10 IOPS to 20 IOPS. Conversely, the IOPS of volume B drops from 50 IOPS to 40 IOPS, although we do not want to do so. 3. Resource Allocation Between Normal Processing and Backup / Copy Processing In a conventional storage system, a case is considered in which a plurality of logical volumes exist on the same disk device and backup and copy are performed in units of individual logical volumes. Conventionally, in order to suppress the influence of the backup / copy processing on normal input / output, a method of setting the pace (interval) of the backup / copy processing at the time of executing the backup / copy processing is adopted.

However, if copying is performed to volume B on the same disk device as volume A while copying volume A, two multiplex copy processes operate on the same disk device at the same time. The effect on the output is doubled. 4. Resource allocation between normal processing and rebuilding In a RAID device, data is made redundant by a plurality of disk drives, so that even if a failure occurs in one disk drive, data can be recovered from the remaining disk drives. For this reason, the RAID device can continue normal input / output even if a failure occurs in the disk drive.

[0011] Further, data recovery is performed on the replaced disk drive from the remaining disk drives. This recovery process is called Rebuilding
Call. Since the rebuilding involves input / output processing for the disk drives constituting the RAID device, the same disk drive competes with normal input / output.

For this reason, the normal input / output performance is reduced by the rebuilding. For example, RA with mirror configuration
In the case of ID1, rebuilding is a process of copying data from a single disk drive due to a disk drive failure to a new disk drive that has been replaced, and read input / output occurs to the copy source disk drive. This read input / output causes the normal input / output to wait, and the performance of the normal input / output deteriorates.

There are two conventional approaches to solving this problem. The first approach copies sufficiently small data at sufficiently long intervals so as not to affect normal I / O. In this case, the influence on the normal input / output can be reduced, but the time until the completion of the rebuilding becomes longer. For example, in the case of RAID 1 composed of a 9 GB disk drive, about 10 hours are required.

A second approach schedules rebuilding I / O if the disk drive is free, ie, not using the disk drive for normal I / O. The problem in this case is that the time to complete the rebuild cannot be guaranteed. This can take a long time to rebuild if the disk drive is almost empty. 5. Maximum Response Time Guarantee In mission-critical tasks, the maximum response time is important in addition to the average response time as a requirement for input / output performance.
Recent disk units have a re-ordering function (Re-o
rdering function).

In the reordering function, the disk device selects the input / output that minimizes the positioning time defined by the sum of the seek time and the rotation wait time from the input / output to be executed as the next input / output. Function. When requesting an input / output from the disk device, the disk device is notified of a simple task (Simple task) for specifying a task to be reordered.

In the case of an input / output designated by a simple task, the disk device schedules the input / output in such an order as to minimize the positioning time. Thereby, the average processing time at the time of random access is reduced. For example, the average processing time of random access is reduced from 9 ms to 5 ms by using the reordering function.

Although the reordering function improves the throughput of the disk device in this way, there is a problem that the maximum response time becomes longer. This is because an input / output for which the positioning time is minimized for the next input / output is selected, so that a phenomenon occurs in which a certain input / output is waited for a long time and is not scheduled.

In order to solve this development, the disk device requires an ordered task (ordered task) in addition to the simple task for specifying that the reordering can be performed.
sk). When an input / output is requested by specifying the ordered task, the disk device completes all the input / output that has been received and has not been completed, and then schedules the input / output of the ordered task.

As described above, by mixing the ordered tasks during the simple task, it is possible to suppress the extension of the maximum input / output response time. However, between random access and sequential access, between logical volumes,
When considering resource allocation between normal processing and backup / copy processing or rebuilding processing, throughput (IOPS)
In addition to using simple tasks to improve the performance, guaranteeing the maximum response time when using simple tasks is an issue.

In order to solve such a problem, the present inventor has proposed a disk time sharing apparatus and a disk time sharing apparatus which can guarantee the minimum performance when a plurality of different types of input / output compete with each other. A method has been proposed (Japanese Patent Application No. 11-218575).

This disk time-sharing device is obtained by grouping a disk device having one or a plurality of disk drives, an input / output request unit for issuing an input / output request to the disk device, and an input / output source to the disk device. An I / O group is formed, and a ratio of time during which each I / O group uses the disk device is defined. Quantum τ1, τ2, at which each I / O group can continuously use the disk device 16 based on the defined time ratio. If τ3 (assigned time) is determined and an I / O request is received from a plurality of I / O groups to the disk device, the disk device is used by sequentially switching the quantum τ1, τ2, τ3 between competing I / O groups. It has an input / output schedule mechanism for performing time sharing.

More specifically, the input / output schedule mechanism includes:
The input / output determined to be sequential access corresponds to the sequential access input / output group, and the other inputs / outputs correspond to the random access input / output group, and time sharing of the disk device is performed by sequential access and random access.

Therefore, no matter how many random access requests occur, the time during which the disk device can be used for sequential access input / output is guaranteed, so that the minimum value of the sequential access performance can be guaranteed. Also, since the time during which the disk device can be used for random access input / output is guaranteed, the minimum random access performance can be guaranteed.

By the way, it is assumed that there is a request from a user who is a system administrator to suppress the average response time of random access input / output to, for example, 30 ms or less. Here, due to disk time sharing, 1
Time sharing cycle TS = 100m
s, the ratio of time between random quantum and sequential quantum (hereinafter referred to as “RS ratio”) is RS ratio = 90%
(However, it is assumed that the time sharing process is performed at a ratio RS = R / TS viewed from the random side).

When the time sharing cycle TS is set to 100 ms under a normal light load condition, the average response time Ave [ms] and the maximum response time Max [ms] are both short. convenient. However, when the load becomes heavy, the processing cannot be completed in the time sharing cycle TS = 100 ms, and the average response Ave is deteriorated.

Therefore, by setting the time sharing period TS to a long time such as 300 ms in advance,
The deterioration of the average response Ave when the load becomes heavy can be suppressed, while the response time when the load is light is T
There is a problem that the response becomes longer than the setting of S = 100 ms, and a quick response cannot be obtained.

It is an object of the present invention to provide a disk time sharing apparatus and a method capable of automatically adjusting operating conditions of time sharing satisfying performance required by a user based on actual values.

[0028]

FIG. 1 is a diagram illustrating the principle of the present invention. First, the present invention is directed to a disk time sharing device including a disk device 16, an input / output request unit 18, and an input / output schedule mechanism 20, as shown in FIG.

Here, the disk device 16 includes one or a plurality of disk drives, and the input / output request unit 18 issues an input / output request to the disk device. Further, the I / O schedule mechanism 20 forms an I / O group in which I / O sources to the disk device are grouped, defines a ratio of time during which each I / O group uses a disk, and based on the defined time ratio. The allocation time (quantum) during which each I / O group can continuously use the disk device is determined, and when an I / O request is received from a plurality of I / O groups to the disk device, the allocation is performed among the competing I / O groups. Time sharing is performed by switching the time in order and using the disk device.

The present invention for such a disk time sharing apparatus is characterized in that a tuning unit 52 for automatically adjusting the operating conditions of the time sharing according to required performance and performance is provided.

For this reason, the disk time sharing apparatus of the present invention stores the results (statistical information) such as the average response, the maximum response, and the throughput obtained by simulation or actual measurement, and stores the load state and the storage by the tuning unit. The optimum adjustment value that satisfies the performance required by the user is determined based on the results of the performance, and the operating condition of time sharing can be automatically adjusted based on the adjustment value, and the performance requirement of the user can be appropriately responded.

Here, the input / output scheduling mechanism 20 forms at least a random access input / output group and a sequential access input / output group as a plurality of input / output groups.

As shown in FIG. 1B, the tuning unit 52 includes a required performance setting unit 56 and first to third basic data 6.
2, 64, 66, and an operating condition determining unit 58. The required performance setting unit 56 includes (1) load state IOPS (actually measured value or set value), (2) average response time Ave [ms] and maximum response time Max [ms] for the random access input / output group, and (3) Each of the throughputs ThP [MB / s] of the sequential access input / output group is set as a required performance value.

The first basic data 62 represents the actual value of the average response Ave divided for each random access load IOPS, the time sharing period TS, and the RS which is the allocated time ratio (quantum ratio) between random access and sequential access. Store it corresponding to the ratio.

The second basic data 64 indicates the actual value of the maximum response Max divided for each random access load IOPS by the time sharing cycle TS and the ratio of the allocated time of the random access to the sequential access RS
Store it corresponding to the ratio.

The third basic data 66 includes a throughput Th
The actual value of P is stored in correspondence with the time sharing period TS and the RS ratio, which is the ratio of the allocated time between random access and sequential access.

Further, the operating condition determining unit 58 calculates a time shelling period TS that satisfies one or a plurality of required performance values set by the required performance setting unit, and an allocation time ratio (RS ratio) between random access and sequential access. An adjustment value is determined by referring to the random load and the first to third basic data, and the time sharing operating condition is automatically adjusted.

If there is a required performance value that cannot be achieved, the tuning unit 52 assigns a priority to the type of required performance,
Automatically adjust operating conditions using one of the following modes: (1) A first mode for determining an adjustment value without considering lower required performance when lower required performance cannot be achieved within a setting range capable of achieving higher priority required performance. (2) The second mode in which the adjustment value is determined in consideration of the lower required performance even when the lower required performance cannot be achieved within the setting range where the higher priority required performance can be achieved. (3) When the lower required performance cannot be achieved in the setting range in which the required performance of the higher priority can be achieved, a third adjustment value for selecting the optimum value of the lower performance from the upper setting range is selected.
mode. (4) When the lower required performance cannot be achieved in the setting range in which the required performance of the higher priority can be achieved, a plurality of candidates having better performances are selected from the higher setting range, and the A fourth mode for selecting an adjustment value at which a higher-level performance is best from the above.

As described above, the priority is assigned to the required performance and the adjustment value is determined, so that even if not all the required performances are satisfied, the automatic adjustment is performed so as to satisfy the high-priority required performance emphasized by the user. Therefore, the user's request can be appropriately reflected.

Further, the present invention provides a disk device having one or a plurality of disk drives, an input / output request unit for issuing an input / output request to the disk device, and a schedule for using the disk device based on the input / output. A disk time sharing method provided with an I / O schedule mechanism that forms an I / O group in which I / O sources to a disk device are grouped and a time during which each I / O group uses a disk. The I / O group determines the allocation time (quantum) that each I / O group can use the disk device continuously based on the defined time ratio, and accepts I / O requests from multiple I / O groups to the disk device. Time, switch the allocation time between competing I / O groups in order to use disk units. Perform sharing, further characterized by automatically adjusting the operating conditions of the time-sharing in response to the required performance and results.

The details of this method are basically the same as those of the apparatus.

[0042]

FIG. 2 is a block diagram of a storage system to which the present invention is applied. In FIG. 2, the storage system includes a device control device 12, an array disk device 14, and a disk device 16.
For the device control device 12, the hosts 10-1 to 10-1
0-n are connected, and input / output requests are sent to the device control device 12 by applications of the hosts 10-1 to 10-n.

The array disk unit 14 receives an input / output request from the device control unit 12 and
Issue the input / output request received for the request. The disk time sharing device of the present invention includes an input / output request requesting unit 18 and a disk input / output schedule mechanism 20 provided in the array disk device 14, a disk input / output processing unit 22 and a disk drive 24-provided in the disk device 16.
1 to 24-n.

When a plurality of disk drives 24-1 to 24-n provided in the disk device 16 have a RAID configuration, the array disk device 14 is further provided with a RAID control mechanism.

Further, in the disk time sharing apparatus of the present invention, a tuning mechanism 50 is provided for the disk input / output schedule mechanism 20 of the array disk apparatus 14, and the tuning mechanism 50 includes a tuning unit 52 and a basic data file 54. .

The tuning unit 52 automatically adjusts the operating conditions of time sharing in the disk input / output scheduling mechanism 20 based on the load and the basic data so as to satisfy the required performance desired by the user.

FIG. 3 is a block diagram of a basic embodiment of the time sharing apparatus of the present invention applied to the storage system of FIG. 2, and exemplifies a disk device having a RAID configuration.

In FIG. 3, the array disk device 14
Includes an input / output request unit 18, a RAID control unit 26, and a disk input / output schedule mechanism 20. The disk device 16 is provided with a disk input / output processing unit 22. Two disk drives 24-1 and 24-2 having a RAID1 configuration (mirror disk configuration) are connected to the disk input / output processing unit 22, for example. Have been.

The disk time sharing apparatus according to the present invention forms an input / output group by grouping input / output requests from the input / output requesting unit 18 to the disk device 16 and forms a disk group. Device 16
Defines the ratio of time to use, and based on the defined time ratio, determines the quantum (allocation time) in which each I / O group can use the disk device continuously, and accepts requests from multiple I / O groups. In this case, a process is performed in which the quantum is sequentially switched between the conflicting input / output groups and the disk device 16 is used.

When there is an input / output request from only one input / output group, scheduling is performed so that the disk device 16 can be continuously used for input / output from one input / output group.

The configuration and function of each unit in FIG. 3 for realizing such a disk time sharing process of the present invention will be described in more detail as follows. Input / output request unit 18
Issues an I / O request to the disk device 16 to the disk I / O schedule mechanism 20 via the RAID control unit 26 based on, for example, a command from the upper device control device 12 shown in FIG. The RAID control unit mainly performs a process of converting the requested logical input / output request into a physical input / output request.

The disk input / output scheduling mechanism 20 includes disk time sharing control information 30-1,
30-2, an input / output schedule unit 32, an input / output request receiving unit 34, and an input / output completion processing unit 36 are provided. Disk time sharing control information 30-1, 30-2
Is the disk drive 2 provided in the disk device 16
4-1 and 24-2 units are provided.

The input / output schedule section 32 includes a disk drive provided for each of the disk drives 24-1 and 24-2.
The disk time sharing is performed by referring to and updating the time sharing control information 30-1 and 30-2.

Here, the disk time sharing control information 30-1 will be described. In this embodiment, a case where the input / output groups are defined as three groups G1, G2 and G3 is taken as an example. Output group G1
Schedule queue group queue 38-corresponding to G3
1, 38-2 are provided. In the schedule waiting group queues 38-1 to 38-3, the I / O requests received by the I / O request receiving unit 34 are assigned to the FIFs constituting the queues.
Lined up by storing in O.

The completion waiting group queues 40-1, 40-2, 40-3 corresponding to the input / output groups G1 to G3.
Is provided. Completion waiting group queues 40-1 and 40
At -3, the input / output request to the disk device 16 is completed,
The input / output requests that have not received the input / output completion response from the disk device 16 are arranged by being stored in the FIFO constituting the queue.

Further, group quanta 42-1, 42-2 and 42-3 are provided corresponding to the input / output groups G1 to G3. Quantum 42-1 to 42 for this group
-3, the input / output groups G1 to G3 are the disk devices 1
6 are defined in advance, and based on the defined ratios α1, α2, α3, the allocation time during which the respective input / output groups G1 to G3 can continuously use the disk device is determined. Quantum τ1, τ
2, τ3 are determined and stored.

For example, assuming that the time sharing cycle for performing one time sharing is TS, the quantum τ1 to τ3 of the input / output groups G1 to G3 is defined by the following equation.

Τ1 = α1 · TS τ2 = α2 · TS τ3 = α3 · TS Disk device 1 of such input / output groups G1 to G3
The appropriate value of the quantum τ1 to τ3 that determines the use of No. 6 is determined as follows. First, if the quantum is set too low, the input / output processing time of the disk device 16 will be close to the value, and the reordering effect of selecting the input / output to minimize the positioning time will be reduced, and the overall input / output performance will decrease I do.

On the other hand, if the value of the quantum is too large, the waiting time of the quantum for switching to another input / output group increases, so that the average input / output processing time and the maximum input / output processing time increase. For example, if the quantum τ1 and the quantum τ2 are set to one hour, respectively, the input and output of the quantum τ2 cannot be performed during the processing of the quantum τ1, so that the input and output of the quantum τ2 waits for one hour until the quantum τ1 ends.

According to the experiment performed by the inventor of the present invention, in the case of the disk device 16 in which the average input / output processing time is several ms to 20 ms, the quantum value is preferably several tens ms to several hundred ms.

In the disk time sharing according to the present invention, if the quanta ヲ R, which is the allocation time for random access, and the quanta, ヲ S, which is the allocation time for sequential access, the quantum ratio of the two (allocation time not known). Is defined as the RS ratio by the following equation.

RS ratio = R / (R + S) This is a quantum ratio viewed from the random access side. By variably setting the RS ratio as an adjustment value, it is possible to change the allocation time of the random access and the sequential access within the time sharing period TS.

Here, the disk input / output schedule mechanism 2
The input / output grouped by 0 includes, for example, the following grouping.

(1) I / O group for random access (2) I / O group for sequential access (3) I / O group by logical volume (4) I / O group by copy / backup processing (5) RAID rebuilding processing The I / O group is formed by the I / O request requesting unit 1
4 is provided with a sequential access detection mechanism 45, a backup detection mechanism 78, and a rebuilding mechanism 84.

For this reason, in the present invention, for example, four input / output groups G1 to G4 for random access, sequential access, copy / backup processing, and rebuilding processing are formed, and Quantum is set for time sharing.

A plurality of input / output groups may be combined into one. For example, random access and sequential access may be combined into one input / output group, and copy / backup processing may be formed as three independent groups. In this case, for random access and sequential access belonging to the same group, R
It has an allocation time according to the S ratio.

Disk time sharing control information 3
0-1 is provided with a current quantum type 44, a current quantum start time 46, and a next input / output task type 48.
The current quantum type 44 is provided for each of the disk drives 24-1 and 24-2 of the disk device 16.
The identifier of the input / output group using the disk drives 24-1 and 24-2 is set.

The current quantum start time 46 is provided for each of the disk drives 24-1 and 24-2 of the disk device 16, and the time T0 at which the current quantum set in the current quantum type 44 starts is set. Further, the next input / output task type 48 is provided for each of the disk drives 24-1 and 24-2 of the disk device 16, and determines whether the input / output request for the next disk drive is a simple task or an ordered task. Is set. The simple task or ordered task set in the next input / output task type 48 is
This is performed to make full use of the effect of the ordering function.

Here, the reordering function of the disk device 16 minimizes the positioning time given by the sum of the seek time and the rotation time from the input / output waiting for execution for each of the disk drives 24-1 and 24-2. This function selects the input / output to be executed next as the input / output to be executed.

When requesting input / output from a disk device having such a reordering function, when a sample task is specified, the disk drive is notified that the reordering may be performed. The disk drive that receives the input / output specifying this sample task schedules the input / output in such an order as to minimize the positioning time.

However, since the reordering function always selects an input / output with a minimum positioning time, a phenomenon occurs in which a certain input / output is not waited for a long time and is not scheduled. In order to solve this phenomenon, the disk drive has an order task function in addition to the simple task. When an input / output is requested by designating the order task, the disk drive completes all the incomplete inputs / outputs that have been inherited and schedules the input / output of the ordered task. For this reason, by mixing ordered tasks between simple tasks, it is possible to suppress the extension of the maximum input / output response time.

In the disk time sharing process of the present invention, the first input / output after quantum switching is performed by specifying an ordered task and using the disk device 16.
To complete the I / O that has not been completed before the quantum switch, and then execute the I / O for the next quantum. For this reason, a simple task is specified for the second and subsequent inputs and outputs after switching to quantum.

If there is only input / output from one input / output group, the input / output group is repeated while resetting the quantum in order to continue the schedule. In this case, the first I / O immediately after resetting the quantum is requested by the ordered task, and the I / O of the reset quantum is scheduled after all the I / O that has not been completed in the previous quantum is completed. I do.

As a result, it is possible to prevent an increase in the maximum response time when the input / output of a plurality of input / output groups conflicts and when only the input / output of one input / output group is continuous.

FIG. 4 shows an example of a disk time sharing schedule by the input / output schedule section 32 provided in the disk input / output schedule mechanism 20 of FIG.

In FIG. 4, three input / output groups G1
ＧG3, the schedule waiting group queue 38-1 of the disk time sharing control information 30-1
38-3, the input / output requests are stored in the group G according to the quantum holding times τ1, τ2, τ3 determined for the input / output groups G1 to G3.
The input / output is scheduled in the order of 1 to G3, and the input / output is requested to the disk device 16.

For example, during the quantum holding time τ1 from time t0, two inputs / outputs of the input / output group G1 are scheduled. In the quantum switching, when the time at the time of completion of the input / output exceeds the current quantum switching time, the quantum is switched to the quantum of the next input / output group. This switching is determined by the following equation. (Current quantum input / output start time) <(current quantum start time + quantum) (1) That is, if the expression (1) is satisfied, the input / output of the input / output group G1 corresponding to the current quantum type is requested to the disk device. If not, it switches to the next input / output group G2 quantum.

During the quantum holding time τ2 of the next input / output group G2, for example, six inputs / outputs are scheduled. Further, when the quantum holding time τ2 elapses at the time t2, the input / output group G3 is switched to the quantum holding time τ3.
Are scheduled. In the same way, the quantum holding times τ1, τ2, τ3 are similarly switched to schedule the input / output of each input / output group.

FIG. 5 shows an example of the time sharing process when only the input / output of a specific input / output group is continuous. In FIG. 5, at time t0, only the input / output of the input / output group G1 is changed to the schedule waiting group queue 38 of FIG.
-1 and the schedule waiting queues 38-2 and 38-3 of the remaining input / output groups G2 and G3 are assumed to be empty.

In this case, the input / output group G1 has a quantum holding time τ1 of the input / output group G1 from time t0.
After the two input / outputs are scheduled, the quantum holding time τ1 is reset at time t1, thereby restarting the quantum 1 holding time 1 of the next same input / output group G1, and, for example, three inputs / outputs are scheduled.

As described above, when only the input / output of one input / output group is in the waiting state, the input / output of one input / output group is continuously scheduled by resetting the quantum.

Further, in FIG. 5, when the inputs and outputs of the three input / output groups G1 to G3 enter a conflicting state at time t2, switching to the next quantum holding time τ2 is performed. However, there are only three inputs / outputs in the input / output group G2 in the quantum holding time τ2, and three input / output requests are interrupted at time t3 in the quantum holding time τ2.

In this case, for example, the input / output group G3
Since there is an I / O request in the waiting state, the time is switched to the quantum holding time τ3 at time t3, and the I / O group G3
For example, schedule three inputs and outputs.

In the disk time sharing schedule shown in FIGS. 4 and 5, the input / output request to the disk drive is made by an ordered task immediately after the quantum is switched, and the second and subsequent times I / O until the next quantum change is requested by a simple task.

As described above, the disk drives 24-1 and 24-2
In order to take advantage of the re-ordering function of 4-2, when the quantum is switched, the current disk drive 24-
Predict the time until all the input / output requests requested to 1,24-2 are completed, and if the predicted time is within the quantum after switching, request quantum input / output after switching, and estimate the time. If the quantum after the switch is exceeded, the input / output after the switch is not requested, and the switching to the next quantum is waited.

This is to create an environment in which the disk I / O schedule mechanism 20 requests the disk unit 16 for as much I / O as possible in order to benefit from the reordering of the disk unit 16.

When the simple task is used, a plurality of input / output requests are requested to the disk device. Since the disk time-shelling of the present invention aims at time-sharing control of the input / output processing time in the disk device, when requesting an input / output request to the disk device, a plurality of requested requests are processed by the disk device. It is necessary to predict the time required to perform the switching, and after switching to the next quantum, determine whether or not to input / output the quantum type after switching to the disk device.

Therefore, the remaining time τr is calculated by the following equation to determine whether the request currently requested to the disk drive at the time of quantum switching is completed within the next quantum and a new input / output request can be input. I do. τr = T0 + τ−Tw−Tnow (2) where T0 is the quantum start time (predicted value), τ is the quantum allocation time Tw is the unprocessed I / O processing time (predicted value), and Tnow is the current time T0 = Ts + Tw (3) Where Ts is the quantum start time before switching Tw = N × Ta (4) where N is the number of unprocessed I / Os Ta is the average I / O processing time for each access type. Means that an I / O request is input to the disk device and no completion response is returned. In the embodiment of the present invention, the unprocessed I / O includes a pre-processed I / O.
O, there are unprocessed I / Os before and all unprocessed I / Os, respectively, the unprocessed I / O of the immediately preceding quantum, the unprocessed I / O of the previous quantum, and the unprocessed through all quantums. I / O.

The quantum start time T0 is also a predicted value, and is predicted when the switching to the current quantum is determined based on the remaining time prediction of the previous quantum. At this time, the time Tw required to complete the processing of all unprocessed I / Os of the previous quantum on the disk device is predicted by equation (4), and the end time of the previous quantum on the disk device, that is, the current time, is calculated. The start time T0 of the quantum is predicted by equation (3).

When the disk input / output scheduler receives a new input / output, the remaining time Tr in the equation (2)
Alternatively, it is calculated when an input / output completion response is received from the disk device. If the remaining time Tr is Tr> 0, it is determined that there is a remaining time, and the current quantum input / output is input to the disk device. If Tr ≦ 0, it is determined that there is no remaining time, and the quantum is switched.

Here, the average input / output processing time T of the disk drive used for calculating the remaining time Tr in the above equation (2)
The calculation method of a is, for example, an average value of the immediately preceding n input / output processing times. In this case, n may be a finite value of, for example, n = 10, or may be, for example, n = ∞, that is, all inputs and outputs from the start of the system.

Further, the calculation of the average value of the input / output processing time can be either a method of calculating the average value for each input / output group or a method of calculating the average value of all the input / output groups.

On the other hand, when a large amount of data is accessed, the positioning time is shorter than the data transfer time. Therefore, the average input / output processing time Ta is estimated from the amount of data to be accessed and the transfer capability of the disk drive. in this case,
The positioning time depends on how much the benefit of the reordering function can be obtained, that is, the number of I / Os to be reordered in the disk drive at that time, the distribution of addresses of individual I / O requests, etc. In the case of a large amount of data access, the ratio of the positioning time to the processing time is small. In this case, the processing time is estimated as (average positioning time) + (data transfer time).

For example, when accessing 1 MB of data with a disk drive having a transfer speed of 20 MB / s, an average rotation waiting time of 3 ms, and an average seek time of 5 ms, the transfer time is 52 ms while the average positioning time is 8 ms.
Therefore, the processing time is 60 ms, which is the sum of the two.

FIG. 6 shows an example of prediction of the remaining time at the time of quantum switching. FIG. 6 shows a case where sequential quantum and random quantum are alternately repeated.
This is an example in which time elapses from (A) to (J).

FIGS. 6A and 6B show an example of predicting the next quantum start time T0 when switching from random quantum to sequential quantum. It is assumed that the remaining time of the random quantum is insufficient at the current time Tnow at which the random quantum has been switched. At this time, the previous quantum sequential I / O requests 1
The random I / O of the current quantum has not returned a completion response of three requests, and is being processed by the disk device.

In this case, as shown in FIG. 6B, the time Tw1 until all the processing I / Os input to the disk device are completed is predicted by the equation (4). The start time T0 of the sequential quantum is determined. In addition, the quantum is switched to the sequential quantum.

FIGS. 6C to 6F show remaining time predictions.
This is an example in which it is determined that there is remaining time. After switching to the sequential quantum in FIG. 6B, the disk I / O scheduling mechanism changes the sequential I / O to the current time T.
Assume that one request has been received by now. At this time, there is one random I / O request as an unprocessed I / O for which a completion response has not been returned in the I / O request requested to the disk device. Immediately, the disk device uses the random I / O of the previous quantum.
One request of O is being processed.

In this case, as shown in FIG. 6E, the time T until one request for random I / O is completed by the disk device is obtained.
w2 is predicted by the equation (4), and the remaining time Tr2 is obtained as shown in FIG. 6F from the equation (2) using the quantum start time T0 obtained in FIG. 6B. In this case, since Tr2> 0, the sequential I / O can be input to the disk device.

FIGS. 6G to 6J show remaining time predictions.
This is an example in which it is determined that there is no remaining time. Time goes further,
It is assumed that the disk input / output scheduling mechanism has received one sequential I / O request at the current time Tnow in FIG. At this time, there is one sequential I / O request as an unprocessed I / O for which a completion response has not been returned in the I / O request requested to the disk device. Immediately, the disk device is processing one request for the sequential I / O of the current quantum.

In this case, as shown in FIG. 6 (I), the time Tw3 until one sequential I / O request is completed by the disk device is predicted by equation (4), and the quantum calculated by FIG. 6 (B) is obtained. Using the start time T0, the remaining time T is obtained from equation (2).
r3 is obtained as shown in FIG. In this case, since Tr3 <0, it is determined that there is no remaining time, and the next random
Switch to quantum.

FIG. 7 is a flowchart of the disk time sharing control processing of the present invention by the input / output schedule unit 32 provided in the disk input / output schedule mechanism 20 of FIG.

The disk time sharing control processing by the input / output schedule unit 32 is performed by the input / output request receiving unit 34 when a certain input / output request is received from the input / output request unit 18 or the input / output completion processing unit. It operates in response to a call from the time when there is a completion report for the input / output requested to the disk device 16 at 36.

First, as shown in the schedule of FIG.
In the disk input / output schedule mechanism 20 of FIG.
A case will be described in which the quantum is sequentially switched among the three competing input / output groups to perform time sharing of the disk drive 24-1.

In step S1, the schedule waiting group queue 38-1 corresponding to the quantum identifier i = 1 set for the current quantum type is checked to determine whether there is a waiting input / output.

Schedule waiting group queue 38-1
If there is a waiting input / output, the process proceeds to step S2, and it is checked whether there is an incomplete input / output in the quantum two months before.
Now, assuming that the quantum identifier i = 1 is the first schedule, since there is no incomplete input / output in the quantum two months before, the process proceeds to step S3, and the remaining time Tr is predicted from the equation (2).

Subsequently, at step S4, the remaining time Tr becomes Tr.
It is checked whether or not> 0, and if this condition is satisfied, it is determined that there is a remaining time, and the routine proceeds to step S8. Step S
4, the first input / output of the current quantum schedule waiting group queue 38-1 is transmitted to the disk drive 2 via the disk input / output processing unit 22 of the disk device 16.
4-1 and set the task of the next input / output task type information 48 to the simple task.

Then, returning to step S1, it is checked whether or not there is a waiting input / output in the schedule waiting group queue 38-1 of the current quantum.
2. The processes of S3 and S8 are repeated. According to the input / output schedule of the input / output group G1 during the quantum holding time τ1, Tr ≦ 0 in step S3, and if it is determined that there is no remaining time, the process proceeds to step S5 to wait for the schedules of the other input / output groups G2, G3. It is checked whether there is a waiting input / output in the group queues 38-2 and 38-3.

At this time, if there is an input / output waiting in the schedule waiting group queue 38-2 of the next input / output group G2, the process proceeds to step S10, and the next input / output group G2 is switched to the quantum holding time τ2. For the type information 48, the next task is set to ordered. At the same time, the quantum current time T0 is predicted from equation (3), and the predicted T0 is set as the current quantum start time.

As a result, the switching from the quantum holding time τ1 of the first input / output group G1 to the quantum holding time τ2 of the next input / output group G2 is performed, and the process returns to step S1 again. G2 is processed by steps S2, S3, S4 and S8.

At this time, since the next task is set to ordered in step S10, the first input / output after the quantum switching is performed by designating the ordered disk drive 24.
-1. When the request is completed, the task of the next input / output task type information 48 is simply set.

Next, a description will be given of a process when input / output requests of one input / output group, for example, the input / output group G1 are continuous as shown in FIG. If the input / output requests of the same input / output group G1 are continuously received, steps S1 to S1 are executed.
The process of S4 is repeated at the quantum t1 to request an input / output request of the same input / output group to the disk drive 24-1. It is checked whether or not there is a waiting input / output in the schedule waiting group queues 38-2 and 38-3.

At this time, if there is no waiting input / output in the schedule waiting group queues 38-2 and 38-3 of the other input / output groups G2 and G3, the process goes to step S9.
Then, the current quantum holding time τ1 is reset and the next task is set to ordered. At the same time, the current quantum time To is predicted and set to the current quantum start time, and the process returns to step S1. In this case, the current quantum type is left as it is.

For this reason, the next quantum after resetting the current quantum holding time τ1 has the same quantum holding time τ1, and when the input / output request of the input / output group G1 continues, the same quantum τ1 is continued.

On the other hand, as shown after time t2 in FIG. 5, the input requests of the same input / output group G1 are continuously made between times t0 and t2, and the quantum t1 and t2 are continued by the reset, and the remaining two are left by time t2. I / O groups G2 and G
When the input / output request of No. 3 is received and stored in the schedule waiting group queues 38-2 and 38-3, the next input / output group G2 is switched to the quantum holding time τ2.

However, when the schedule waiting group queue 38-2 of the input / output group G2 becomes empty in the middle of the quantum holding time τ2 as shown at time t3 in FIG. Proceeding to S6, it is checked whether or not there is a waiting input / output in the schedule waiting group queues 38-1 and 38-3 for another quantum.

At this time, if there is a waiting input / output in another quantum, the process proceeds to step S10, and it is checked whether all quantums have an uncompleted input / output. If not, the process proceeds to step S11 and the next input / output is performed. Switch to quantum hold time τ3 of group G3, set next task to ordered,
Further, the quantum start time T0 is predicted and set, and by returning to step S1, the first I / O request of the I / O group G3 after switching is ordered to the disk drive 24.
-1 through the processing of steps S1 to S4 and S8.

Here, in the case of forming a group of input / output for sequential access and random access in FIG. 3, a sequential access detection mechanism 45 is provided in the input / output request unit 18. For example, an interface for notifying the input / output of the sequential access detected by the sequential access detection mechanism 45 to the input / output request interface to the RAID control unit 26 is added.

The sequential access detection mechanism 45
The address of the next input / output command is recognized from the address and the data length included in the input / output command issued from the upper device control device 12 shown in FIG. 2, and the address of the next input / output command is predicted. If it is determined that the sequential access is detected, the sequential access is detected, and information such as a flag indicating the sequential access is issued to the input / output request receiving unit 34 of the disk input / output schedule mechanism 20 via the RAID control unit 26 through the interface.

Therefore, the input / output request receiving unit 34 can recognize whether the input / output request received from the input / output request unit 18 is sequential access or random access.

When forming an input / output group for copy / backup processing, a backup mechanism 78 is provided in the input / output request unit 18. The input / output from the backup mechanism 78 is notified to the RAID controller 26 by an additional interface for notifying the backup input / output. The RAID control unit 26 informs that it is a backup input / output at the time of the input / output request to the disk input / output schedule mechanism 20, and performs disk time sharing for the input / output group of the copy processing / backup processing.

Further, when forming an input / output group for the rebuilding process, the input / output request unit 18 is provided with a rebuilding mechanism 84. At the time of the rebuilding process, an interface indicating that the input / output request to the RAID control unit 26 is the rebuilding process is added. When requesting input / output from the disk input / output schedule mechanism 20, the RAID control unit 26 notifies the input / output of the rebuilding input / output and performs disk time sharing for the input / output group of the rebuilding process.

FIG. 8 is a functional block diagram of the tuning mechanism 50 of FIG. In FIG. 8, the tuning mechanism 5
0 is composed of a tuning unit 52 and a basic data file 54. The tuning unit 52 includes a required performance setting unit 56 and an operation condition determining unit 58.

The required performance setting unit 56 calculates the average response Ave and the maximum response M of random access by the user.
ax and sequential access throughput ThP
The random access load state IOPS observed by the disk input / output schedule mechanism 20 of the array disk device 14 shown in FIG. 2 is obtained and output to the operating condition determination unit 58. The load state I of random access
The OPS may be directly supplied to the operation condition determining unit 58.

The operating condition determining unit 58 is configured to set the required performance setting unit 5
The time sharing period TS that satisfies the required performance value set in 6 and the RS ratio of the quantum ratio of random access and sequential access are determined as adjustment values.
Automatically adjusts the time sharing cycle TS of the disk input / output scheduling mechanism 20 and the quantum of each group.

The basic data file 54 includes first basic data 62 for average response and second basic data 62 for maximum response.
Basic data 64 and third basic data 6 for throughput
6 is stored.

FIG. 9 shows the data structure of each basic data stored in the basic data file 54 of FIG. The basic data stored in this data structure is data obtained by simulation or actual measurement.

FIG. 9A shows the first basic data 62 relating to the average response of random access, which stores the corresponding basic data for each random access load IOPS. For example, load IOPS = 100,1
For 50, basic data on the average response is stored.

Taking the load IOPS = 100 as an example, the time sharing period TS = 100 ms, 200 ms, 3
The average response time corresponding to the combination of 00 ms and RS ratio = 90%, 80%, 70% is stored as basic data. Similarly, for load IOPS = 150, an average response is stored corresponding to a combination of three time sharing periods TS and three RS ratios.

FIG. 9B shows the second basic data 64 relating to the maximum response of the random access.
Similar to the basic data 62 of the average response in (A), the basic data is stored separately for random access loads IOPS = 100 and 150. Each basic data has a time sharing cycle TS = 100 ms, 200 ms,
The maximum response is stored as basic data corresponding to the combination of 300 ms and RS ratio = 90%, 80%, 70%.

FIG. 9C shows the third basic data 66 related to the sequential access throughput ThP. Regarding this throughput, regardless of the random access load IOPS, the time sharing period TS
And the throughput corresponding to the combination of RS and RS ratio is stored as basic data.

Next, a description will be given of the tuning processing by the operating condition determining unit 58 in FIG. 8 with reference to the basic data in FIG. 9 as an example. First, it is assumed that the priority is instructed from the required performance setting unit 56 in the order of the average response, the maximum response, and the throughput, and the request value of the user at this time is the following value. -Average response Ave = 40 ms or less-Maximum response Max = 80 ms or less-Throughput ThP = 3.0 MB / s or more Also, assume that the observed value of the random access load state at this time is 100 IOPS.

FIG. 10 shows the procedure of the tuning process in a state where such required performance and priority are set. First, regarding the average response having the highest priority, the load IOPS = 1 in the first basic data 62 in FIG.
00 is extracted as the first basic data 62A in FIG. 10, and a hatched area where the average response Ave is equal to or less than the required performance value of 40 ms is extracted from the first basic data 62A.

Next, data of random access load IOPS = 100 is extracted from the second basic data 64 of FIG. 9B as the second basic data 64A of FIG. 10, and the maximum response 80 ms requested by the user is 80 ms. Obtain the shaded area where you can achieve: Subsequently, the shaded portion of the first basic data 62A and the second basic data 64A are output by the common area inspection unit 71.
In the comparison of the hatched portions, the area of the hatched portion indicated by the first common data 68 that achieves both the average response and the maximum response user request is obtained.

Next, with respect to the third basic data 66 relating to the throughput of FIG.
An area in which the user request of the throughput of B / s or more can be achieved is obtained as shown by the shaded portion of the third basic data 66 in FIG.

Finally, the hatched area where the user request of the average response and the maximum response of the first common data 68 is achieved, and the hatched area where the user request of the throughput of the third basic data 66 is achieved As a result, the common portion of the area where all the user requests of the average response, the maximum response, and the throughput can be achieved is obtained as shown by the shaded portion of the second common data 70.

From the above results, the time sharing period TS = 30 corresponding to the shaded area of the second common data 70
The combination of 0 ms and RS ratio = 90% is the average response,
The maximum response and the throughput are determined as adjustment values that can achieve all user requests, and are set in the disk input / output scheduling mechanism 20 in FIG. 2 to automatically adjust the operating conditions of time sharing.

For example, disk input / output schedule mechanism 2
0, when time sharing of two groups of random access and sequential access is performed, the quantum of random access is set to 270 ms based on the RS ratio = 90% at the same time as setting the time sharing cycle TS = 300 ms. The quantum of the sequential access is set to 30 ms.

FIG. 11 is a flowchart of the tuning process of FIG. First, in step S1, the basic data of the required performance is acquired in the order of the priority of the user request.
For example, when the priorities are set in the order of the average response, the maximum response, and the throughput, first, basic data on the average response is obtained.

Subsequently, in step S2, an area in which setting that can achieve the user request value can be set is obtained for the basic data. Subsequently, if there is a request that can satisfy the user request in step S3, the process proceeds to step S4, and it is determined whether or not the basic data currently being processed is an item of the highest priority required performance.

If the requested performance has the highest priority, the common area cannot be determined because it is the first data area. Therefore, the process returns to step S1 to acquire the requested area for the next required performance. This is performed in steps S1 to S3.
If the requested performance has the second or higher priority, the process proceeds to step S5, and the area of the requested performance that has already been obtained is obtained as a new common area.

Subsequently, it is checked in step S6 whether or not the common area has been obtained. If the common area has been obtained, the flow advances to step S7. In step S7, if there is the next item of the user required performance, the process returns to step S1, and if not, the process proceeds to step S8. In step S8, from the finally obtained common area of the plurality of required performances, a combination in which the item of the highest priority required performance takes the best value is selected.

Here, in the selection of the best combination in step S8, for example, when there is one or more combinations of the time sharing period TS and the RS ratio that satisfy all of the three required performances of the average response, the maximum response, and the throughput. Although there is no problem in the above, when the lower required performance is not achieved, the adjustment value is determined in any one of the first to fourth modes, for example. For example, if the required performance is satisfied for the upper average response and the maximum response but the required performance cannot be achieved for the lower throughput, the processing in modes 1 to 4 is as follows.

(1) When the required performance of the lower throughput cannot be achieved within the setting range in which the average response and the maximum response as the upper priority can be achieved, the adjustment value is determined without considering the lower throughput.

(2) In the second mode, even when the required performance of the lower throughput cannot be achieved within the setting range in which the average response and the maximum response having the higher priority can be achieved, the required performance of the lower throughput is considered. Determine the adjustment value.

(3) In the third mode, when the required performance of the lower throughput cannot be achieved within the setting range in which the required performance of the average response and the maximum response having the higher priority can be achieved, the upper average response and the maximum response can be achieved. From the setting range of the common area, the adjustment value that maximizes the performance of the lower throughput is selected.

(4) In the fourth mode, when the required performance of the lower throughput cannot be achieved within the setting range in which the average response and the maximum response having the higher priority can be achieved, the common area of the upper average response and the maximum response is used. From the set range, a plurality of candidates for which the performance of the lower throughput is improved are selected, and an adjustment value at which the higher average response and the maximum response are maximized is selected from the selected candidates.

FIG. 12 shows a specific example of the adjustment value determination when the lower required performance cannot be achieved within the setting range in which the higher priority required performance can be achieved. Here, the required performance of the user is the same as in the case of FIG. 9:-Average response Ave = 40 ms or less-Maximum response Max = 80 ms or less-Throughput ThP = 3.0 MB / s or more.

The first basic data 6 of the average response
2. The second basic data 64 of the maximum response is shown in FIG.
(A) Same data as (B). On the other hand, the third basic data 66 regarding the throughput is slightly different from the case of FIG. 9C, and is the third basic data 660 of FIG. The difference is that the time sharing period TS = 30
The point is that the throughput is 2.6 MB / s for the combination of 0 ms and RS ratio = 90%.

In the tuning processing shown in FIG. 13, the shaded area of the first basic data 62A extracted from the average response of 40 ms or less and the shaded area of the second basic data 64A extracted from the maximum response of 80 ms or less. The point that the first common data is obtained by the judgment of the common area inspection unit 71 is the same as that of FIG.

On the other hand, with respect to the third basic data 660 of the throughput, the area satisfying the throughput requested by the user is indicated by oblique lines, and the second basic data as the detection result of the first common data 68 by the common area inspection unit 72 is obtained. In the data 720, an area where all the user requests of the average response, the maximum response, and the throughput can be achieved cannot be obtained.

In such a case, since the required performance of the lower throughput is not considered in the mode 1, the first
One of the shaded common areas of the common data 68A is selected. In the second mode, when selecting the shaded common area in the first common data 68A, the third basic data 660 of the lower throughput is taken into consideration, and the common area corresponding to 3.6 MB / s at which the throughput is maximized. Time sharing cycle TS = 300 ms and RS ratio =
Select 90% combination.

In the case of the third mode, the improvement of the throughput is the best among the corresponding areas in the third basic data 660 of the lower throughput corresponding to the three common areas of the first common data 68A. Time sharing period TS = 300 ms corresponding to .6 MB / s and RS ratio =
Select 90%. In this case, mode 3 has the same selection result as in mode 2.

FIG. 13 is an explanatory diagram of the adjustment value selection processing in mode 4 when the lower required performance cannot be achieved in the setting range in which the required performance of the higher priority can be achieved.

In FIG. 13, the time sharing periods TS = 200 ms and 300 ms and the RS ratio = 9 in the third basic data 660 of the lowest priority throughput.
The data stored in the 0% combination is 3.4 MB / s,
3.6 MB / s, which is different from the case of FIG.

In the case of FIG. 13, the request values of the user are as follows: average response Ave = 40 ms or less; maximum response Max = 80 ms or less; throughput ThP = 4.0 MB / s or more. Has become.

Also in this case, as in FIG. 12, there is no area in the second common data 720 in which all user requests of the average response, the maximum response, and the throughput can be achieved. In this case, in the fourth mode, the lower throughput from the third basic data 660 of the throughput corresponding to the three shaded areas of the first common data 68 that has acquired the common area of the average response and the maximum response. A plurality of points are selected as candidates for a region in which the performance of is improved.

In this case, the throughput is 3.4 MB / s
And 3.6 MB / s are selected. From the two candidates selected in this way, the candidate having the highest average response and maximum response performance is selected. That is,
The time sharing period TS = 200 ms and the RS ratio 90 at which the average response is 25 ms in the first basic data 62A and the maximum response is 60 ms in the second basic data 64A
A set of% is selected.

FIG. 14 is a characteristic diagram of a simulation result when automatic tuning is not performed in FIG.

In this simulation, random access, sequential access, OPC access (copy access), EC access (error access)
For each of the four input / output groups, four quanta are set for time sharing. Random access and sequential access belong to the same group. All other OPC accesses and EC accesses belong to independent groups.

In the simulation, since sequential input / output requests are prevented from flowing, all the quanta of sequential access belonging to the same group are used for the quanta of random access.

The time ratio of each quantum is (random): (sequential): (OPC): (E
C) = 65: 5: 15: 15. Since there is no input / output request for the sequential quantum, (Random) :( OPC) :( EC) = 70: 15: 1
5 The time sharing cycle TS is 10
0 ms. Furthermore, random access is 7.5ms
Load 20 IOPS, 100 IOPS, 220I for each
This amplitude is repeated as OPS, 100 IOPS. FIG. 15 is an enlarged view of the simulation start time 0 to 100 ms in FIG.

When the automatic tuning shown in FIGS. 14 and 15 is not performed, the highest 220 IOPS is obtained.
Can't handle the random access load of
It can process only the random access of 200 IOPS like the unit.

When the load is high, the average response (R Ave) of random access is 12
0ms, worst case 150m
Deterioration up to s.

FIG. 16 shows a simulation result when the automatic tuning according to the present invention is performed. FIG. 17 shows an enlarged view of the vicinity of 0 to 100 ms in FIG.

The way of giving the random access load IOPS in this simulation is the same as that of FIG. 14. The time sharing period TS and the time ratio of each quantum are such that the required performance is fixed, but the observed value of the load IOPS. Automatically fluctuates due to In the automatic tuning setting, the average response,
By arranging the maximum response and the throughput in this order, tuning for random access is performed.

As a result, in the case of the automatic tuning shown in FIGS.
Can almost handle the random access load. Also, the average response when the load IOPS is high is about 50 ms as in the part B, and is suppressed to a little over 60 ms as in the part C at worst.

In the above-described embodiment, the average response, the maximum response, and the throughput are taken as examples of the required performance. However, an appropriate required performance can be set as required. Also, the priority is set in the order of the average response, the maximum response, and the throughput by setting the priority to random access. However, the priority may be set in the order of the throughput, the average response, and the maximum response in the sequential access priority. good.

The present invention includes appropriate modifications that do not include the objects and advantages thereof, and are not limited by the numerical values shown in the above embodiments.

[0170]

As described above, according to the present invention, results (statistical information) such as load, average response, maximum response, and throughput obtained by simulation or actual measurement are stored as basic data, and the tuning unit Based on the load condition and the performance based on the stored basic data, the optimal adjustment value that satisfies the performance required by the user, for example, the time sharing cycle and the quantum ratio (RS ratio) of random access and sequential access is determined. By automatically adjusting the operating conditions of the time sharing based on this, it is possible to perform input / output processing appropriately corresponding to the performance required by the user.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is a block diagram of a storage system to which the present invention is applied;

FIG. 3 is a functional block diagram of a basic embodiment of the present invention forming three input / output groups;

FIG. 4 is an explanatory diagram of a schedule of a disk time sharing process for input / output of the three input / output groups of FIG. 3;

FIG. 5 is a diagram illustrating a schedule of a disk time sharing process when input / output of only one input / output group is continuous.

FIG. 6 is an explanatory diagram of a process of estimating a remaining time at the time of quantum switching.

FIG. 7 is a flowchart of a disk time sharing process of FIG. 3;

FIG. 8 is a functional block diagram of the tuning mechanism of FIG. 2;

FIG. 9 is an explanatory diagram of average response, maximum response, and actual throughput values stored as basic data in the basic data file of FIG. 8;

FIG. 10 is an explanatory diagram of the tuning processing of FIG. 8 for selecting an adjustment value according to the priority of required performance.

FIG. 11 is a flowchart of a tuning process in FIG. 8;

FIG. 12 is an explanatory diagram of a tuning process when a lower required performance cannot be achieved;

FIG. 13 is an explanatory diagram of another tuning process when a lower required performance cannot be obtained.

FIG. 14 shows load IO when tuning is not performed.
Characteristic diagram of simulation result of PS, average response and maximum response of random access, copy process, error process execution

FIG. 15 is a partially enlarged view of FIG. 14;

FIG. 16 shows a load IOPS when tuning is performed,
Average and maximum response of random access,
Characteristic diagram of simulation results of execution of copy processing and error processing

FIG. 17 is a partially enlarged view of FIG. 16;

[Explanation of symbols]

10-1 to 10-m: Host 12: Device control device 14: Array disk device 16: Disk device 18: I / O request unit 20: Disk I / O schedule mechanism 22: Disk I / O processing unit 24-1 to 24-n : Disk drive 26: RAID control unit 30-1 to 30-4: Distime shearing control information 32: Input / output schedule unit 34: Input / output reception unit 36: Input / output completion processing unit 38-1 to 38-3: Schedule Waiting group queue 40-1 to 40-3: Completion waiting group queue 42-1 to 42-3: Group quantum 44: Current quantum type information 45: Sequential access detection mechanism 46: Current quantum start time 48: Next input / output task Type information 50: Tuning mechanism 52: Tuning unit 54: Basic data file File 56: Required performance setting unit 58: Operating condition determination unit 62: First basic data 64: Second basic data 66: Third basic data 78: Backup / copy mechanism 84: Rebuilding mechanism

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Masafumi Kato 4-1-1, Kamidadanaka, Nakahara-ku, Kawasaki-shi, Kanagawa F-term within Fujitsu Limited (reference) 5B014 EB04 GD05 GD17 GD23 GD36 HA01 HA04 HA06 HA08 5B065 CA04 CA11 CA16 CA30 CC08 5B082 CA01 EA07 FA16 GA15 JA04

Claims (4)

[Claims] 1. A disk device having one or a plurality of disk drives, an input / output request unit for issuing an input / output request to the disk device, and an input / output group formed by grouping input / output sources to the disk device At the same time, the ratio of time during which each I / O group uses the disk is defined. Based on the defined time ratio, the allocation time (quantum) at which each I / O group can use the disk device continuously is determined. A disk provided with an input / output scheduling mechanism for performing time sharing using the disk device by sequentially switching the allocation time between competing input / output groups when an input / output request is received from the output group to the disk device;・
A disk time sharing device, characterized in that the time sharing device is provided with a tuning unit for automatically adjusting the time sharing operation condition according to required performance and performance. 2. The disk time sharing apparatus according to claim 1, wherein said input / output scheduler forms at least a random access input / output group and a sequential access input / output group as said plurality of input / output groups. A tuning unit configured to set an average response time and a maximum response time of the random access I / O group, and a throughput of the sequential access I / O group as required performance, and a random access load; The first basic data stored in accordance with the time sharing period, the allocated time ratio of random access and sequential access, and the actual value of the average response obtained as described above, and the actual value of the maximum response divided for each load of random access. To Thailand The second basic data stored corresponding to the sharing period and the allocated time ratio between random access and sequential access, and the actual value of the throughput are corresponded to the time sharing period and the allocated time ratio between random access and sequential access. The third basic data stored and stored, a time shelling cycle that satisfies one or a plurality of required performance values set by the required performance setting unit, an allocation time ratio between random access and sequential access, An operating condition determining unit that determines an adjusting value by referring to the first to third basic data and automatically adjusts the operating condition of the time sharing. . 3. The disk time sharing apparatus according to claim 1, wherein the tuning unit assigns a priority to the type of the required performance when there is a required performance value that cannot be achieved, and If the lower required performance cannot be achieved within the setting range that can achieve the required performance, the first mode determines the adjustment value without considering the lower required performance, and the setting range that can achieve the required performance of the higher priority In the second mode, where the lower required performance cannot be achieved, the adjustment value is determined in consideration of the lower required performance, and when the lower required performance cannot be achieved in the setting range in which the higher priority required performance can be achieved. In addition, a third mode in which an adjustment value at which the lower performance is optimized from the upper setting range is selected, and the lower required performance cannot be achieved in the setting range in which the higher priority required performance can be achieved. And a fourth mode for selecting a plurality of candidates for which the lower performance is better from the upper setting range, and selecting an adjustment value for which the higher performance is best from the selected candidates. A disk time-sharing device characterized by automatically adjusting operating conditions by using a computer. 4. A disk device having one or a plurality of disk drives, an input / output request unit for issuing an input / output request to the disk device, and an input / output for scheduling use of the disk device based on the input / output. In a disk time sharing method having a scheduling mechanism, an I / O group is formed by grouping I / O sources to the disk device, and a ratio of time during which each I / O group uses a disk is defined. If the I / O group determines the allocation time (quantum) for which the I / O group can use the disk device continuously based on the defined time ratio, and if I / O requests are received from multiple I / O groups to the disk device, contention occurs. Time sharing in which disk units are used by sequentially switching the allocation time between There, further, disk time-sharing method characterized by automatically adjusting the operating conditions of the time-sharing in response to the required performance and results.