JP2009199367A - Computer system, i/o scheduler and i/o scheduling method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To increase a throughput under a condition that all processes execute disk access at least once within a prescribed time. <P>SOLUTION: An I/O scheduler performs the scheduling of access requests from a plurality of processes. The scheduling includes a step (A) for reading process characteristic information indicating an access target file and file position information, a step (B) for storing an access request in a wait queue, a step (C) for determining the order of wait queues so that the access target file is accessed in the arrangement order on a disk while referring to the process characteristic information and the file position information, a step (D) for determining the number of access requests to be taken out from respective wait queues so that all processes execute disk access at least once within the prescribed time, and a step (E) for acquiring the access requests from the wait queues and sending the acquired access requests to the disk in accordance with the determined order and number of access requests. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a computer system. In particular, the present invention relates to an I / O scheduling technique for scheduling a disk access request in a computer system.

  In a computer system, many processes executed by a CPU require a disk access process (data reading, data writing). Therefore, speeding up the disk access processing is one of the important issues for speeding up the computer system. In order to increase the speed of the disk access processing, it is important to appropriately adjust the order of the disk access requests sent to the disk, instead of directly passing the disk access requests from many processes to the disk. Performing such adjustment (arbitration) is called “scheduling”. The “I / O scheduler” in the OS performs scheduling of disk access requests from a plurality of processes.

  Patent Document 1 discloses an external storage device shared by a plurality of computers. The external storage device includes a plurality of storage disk units for storing data and an external storage control unit. The storage disk unit can queue read / write commands for data. The external storage control unit controls data input / output processing between the computer and the storage disk unit by storing commands from the computer in an acceptance queue and sequentially issuing the commands to the storage disk unit. More specifically, the external storage control unit includes a predicted processing time generation unit, a sending queue, a processing time prediction unit, a command queuing unit, and a command batch generation unit. The predicted processing time generation unit generates a command processing time predicted value that is a predicted time required for command processing. The send queue is a queue of commands. In the sending queue, the commands in the receiving queue are stored at a predetermined timing as many times as the sum of the command processing time predicted values becomes a value corresponding to a predetermined processing time slice. The processing time prediction means predicts the processing time for each storage disk unit based on the predicted processing time generation means. The command queuing means takes out a command for the storage disk unit having the maximum estimated processing time from the transmission queue and queues the command to the corresponding storage disk unit. The command batch generation means stores the commands in the receiving queue in the sending queue when each sending queue becomes empty.

JP-A-9-258907

  In recent years, CPU performance has improved rapidly, but disk performance has not improved as rapidly as CPU. For this reason, there may be a process or CPU in which the waiting time from when a disk access request is issued until a response is received becomes very long. If the waiting time becomes long or the waiting state frequently occurs, the efficiency of the entire system is lowered. Therefore, it is preferable that the I / O scheduler schedules disk access requests from a plurality of processes so that the waiting time is shortened.

  It is also important to improve the disk bandwidth (throughput), which is the data transfer rate per unit time between the disk and the CPU, in order to speed up the disk access process. Therefore, it is preferable that the I / O scheduler schedules disk access requests from a plurality of processes so as to improve the throughput.

  However, in general, “reduction of waiting time” and “improvement of throughput” are in a mutually contradictory relationship. This will be described with reference to the example shown in FIG.

  In FIG. 1, it is assumed that processes P1 to P4 access files F1 to F4 recorded on the disc. The I / O scheduler determines in what order the disk access requests issued from each of the processes P1 to P4 are sent to the disk. More specifically, the I / O scheduler has wait queues Q1 to Q4 that temporarily hold disk access requests. The wait queue is assigned for each process. In the example of FIG. 1, the wait queues Q1 to Q4 are assigned to the processes P1 to P4, respectively. That is, disk access requests from the processes P1 to P4 are temporarily stored in the waiting queues Q1 to Q4. The request selection unit takes out the disk access requests from the waiting queues Q1 to Q4 according to the “predetermined rule”, and sequentially sends the taken out requests to the disk.

  Here, as the predetermined rule, (1) one disk access request is sequentially taken out from each waiting queue, or (2) all disk access requests stored in one waiting queue are taken out. After that, it is possible to move to the next waiting queue. Each of these two rules has advantages and disadvantages.

  In the case of rule (1), the request selection unit takes out one disk access request from a certain waiting queue and immediately moves to the next waiting queue. As a result, after one process accesses a file on the disk, another process immediately accesses another file on the disk. Therefore, each process can access the disk one after another with a short waiting time. Since no process has an extremely long waiting time, no timeout occurs and the processing efficiency of each process is improved. On the other hand, since the files to be accessed on the disk are switched one after another, it is necessary to move the disk head frequently, and it takes time to seek between the files. When the seek time between files increases, the amount of data that can be read (or written) from each file within a certain period of time inevitably decreases. This means that the throughput, which is the data transmission rate per unit time, decreases. Thus, in the case of rule (1), the waiting time is shortened, but the throughput is reduced.

  In the case of rule (2), the request selection unit extracts all disk access requests stored in one wait queue, and then moves to the next wait queue. Therefore, the file to be accessed does not switch frequently. In general, data accessed by the same process exists at a close position on the disk. Accordingly, the seek time of the disk head is shortened, and the throughput, which is the data transfer rate per unit time, is increased accordingly. On the other hand, the next process needs to wait until processing of all disk access requests from a process is completed. In other words, the later process has a longer waiting time and the possibility of a timeout occurring. This tendency becomes stronger as the number of processes operating simultaneously increases and as the number of disk access requests issued by one process increases. Thus, in the case of rule (2), the throughput is improved, but the waiting time is increased.

  The inventor of the present application paid attention to the following points. In actual system operation, the waiting time of each process does not need to be minimum, and in many cases, if each process starts processing within a predetermined time (for example, a time when an application timeout does not occur) Good. In other words, each process may perform disk access at least once within a predetermined time. Under such conditions, it is desired to increase the throughput as much as possible.

  One object of the present invention is to provide a scheduling technique capable of increasing the throughput as much as possible under the condition that all processes execute disk access at least once within a predetermined time.

  In a first aspect of the present invention, an I / O scheduler is provided. When a plurality of processes access a disk, the I / O scheduler causes a computer to execute a scheduling process of a disk access request issued from each process. The scheduling process includes (A) reading process characteristic information indicating an access target file accessed by each of a plurality of processes and file position information indicating a position of a file recorded on the disk from a storage device; B) A step of storing disk access requests from a plurality of processes in each of a plurality of wait queues, and (C) referring to process characteristic information and file position information, and the access target files are sequentially arranged in the order of arrangement on the disk. Determining the order of a plurality of wait queues from which disk access requests are retrieved so as to be accessed, and (D) the predetermined process so that all of the plurality of processes perform disk access at least once within a predetermined time period. The disk is removed from each of the multiple waiting queues within Determining a number of access requests, according to the order and the number determined above (E), comprising the steps of: sending from a plurality of waiting queue to get the disk access request, the disk access request to the disk obtained, the.

  In a second aspect of the present invention, there is provided an I / O scheduling method for performing a scheduling process of a disk access request issued from each process when a plurality of processes access the disk. In the I / O scheduling method, (A) a step of reading process characteristic information indicating an access target file accessed by each of a plurality of processes and file position information indicating a position of a file recorded on a disk from a storage device (B) a step of storing disk access requests from a plurality of processes in each of a plurality of waiting queues; and (C) referring to process characteristic information and file position information, and the access target files are arranged in order on the disk. (D) determining the order of a plurality of waiting queues from which disk access requests are taken out, so that all of the plurality of processes perform disk access at least once within a predetermined time period. And is taken out from each of the plurality of waiting queues within the predetermined time. And determining the number of I disk access request, in accordance with (E) the determined order and number, and sending a plurality of waiting queue to get the disk access request, the disk access request to the disk obtained, the.

  In a third aspect of the present invention, a computer system is provided. The computer system includes a disk, a processing device that executes a plurality of processes that access the disk, an I / O scheduler, and a storage device. The storage device stores process characteristic information indicating an access target file accessed by each of a plurality of processes, and file position information indicating a position of a file recorded on the disk. The I / O scheduler is executed by the processing device and performs a scheduling process of a disk access request issued from each of a plurality of processes. More specifically, the I / O scheduler stores disk access requests from a plurality of processes in each of a plurality of wait queues. The I / O scheduler refers to the process characteristic information and the file position information, and determines the order of a plurality of wait queues from which disk access requests are retrieved so that the access target file is accessed in order in the arrangement order on the disk. . The I / O scheduler determines the number of disk access requests fetched from each of the plurality of wait queues within a predetermined time period so that all of the plurality of processes perform disk access at least once within the predetermined time period. Then, the I / O scheduler acquires disk access requests from a plurality of waiting queues according to the determined order and number, and sends the acquired disk access requests to the disk.

  According to the scheduling technique of the present invention, it is possible to maximize the throughput under the condition that all processes execute disk access at least once within a predetermined time.

  A computer system and an I / O scheduler according to an embodiment of the present invention will be described with reference to the accompanying drawings.

1. Computer System FIG. 2 shows an example of the configuration of a computer system (computer system) 1 according to an embodiment of the present invention. The computer system 1 includes a processing device 2, a storage device 3, a disk 4, an input device 5, and an output device 6. The processing device 2 includes one or a plurality of CPUs. The storage device 3 is, for example, a RAM (Random Access Memory). The disk 4 is, for example, an HDD (Hard Disk Drive) in which a plurality of files are recorded. The disk 4 may be shared by a plurality of computer systems 1 via a network. The input device 5 is, for example, a keyboard or a mouse. The output device 6 is a display, for example. The computer system 1 may be constructed by a virtualization technique.

  The processing device 2 executes various applications in addition to the OS. At this time, the processing device 2 executes a process generated by the OS, application, or the like. Many processes require access processing (data read, data write) to the disk 4 and issue a disk access request to the I / O scheduler 10 in the OS. The I / O scheduler 10 is a kind of program executed by the processing device 2, and performs scheduling processing of disk access requests issued from each of a plurality of processes. The OS and the I / O scheduler 10 are recorded on the disk 4 or a computer-readable recording medium. Then, the OS and the I / O scheduler 10 are read into the storage device 3 and executed by the processing device 2.

  As will be described in detail later, the I / O scheduler 10 according to the present embodiment increases the throughput as much as possible under the condition that all processes execute disk access at least once within a predetermined time. Perform scheduling processing. For this purpose, the I / O scheduler 10 uses process characteristic information PROC, file position information LOC, seek time information SEK, access time information ACS, and the like. The process characteristic information PROC, file position information LOC, seek time information SEK, and access time information ACS are stored in the storage device 3. First, details of each information will be described.

(Process characteristic information PROC)
FIG. 3 shows an example of the process characteristic information PROC. The process characteristic information PROC indicates a file (access target file) on the disk 4 accessed by each process. For example, in the example of FIG. 3, the process characteristic information PROC indicates that the plurality of processes P1 to P4 access the plurality of files F1 to F4, respectively. It can be said that the process characteristic information PROC indicates the correspondence between the processes (P1 to P4) and the access target files (F1 to F4). By referring to the process characteristic information PROC, the I / O scheduler 10 can grasp which access target file on the disk 4 is accessed by each process.

(File location information LOC)
4A and 4B show examples of the file position information LOC. The file position information LOC is information indicating the position of the file recorded on the disk 4. In the example of FIG. 4A, the correspondence between the file name, the start position, and the end position is shown for each file. In the example of FIG. 4B, the correspondence between the file name, the start position, and the file size is shown for each file. In the case of FIG. 4B, the end position of each file can be obtained from the start position and the file size. By referring to the file position information LOC, the I / O scheduler 10 can calculate the arrangement relationship of the files on the disk 4 and the distance between the files. In the example shown in FIGS. 4A and 4B, the number of bytes from the top of the disk 4 is used as a unit representing the file position, but is not limited thereto. The file position may be represented by the number of cylinders from the top of the disk 4.

(Seek time information SEK)
5A and 5B show examples of seek time information SEK. The seek time information SEK is information indicating the correspondence between the “seek distance” that is the distance the head moves on the disk 4 and the “seek time” that is the movement time. FIG. 5A is a graph showing the correspondence y = f (x). In FIG. 5A, the horizontal axis represents the seek distance x, and the vertical axis represents the seek time y. Alternatively, as shown in FIG. 5B, the seek time information SEK may give the correspondence between the seek distance and the seek time in a table format. According to FIG. 5B, for example, when the seek distance is 0.5 GB, the seek time is 8.0 milliseconds, and when the seek distance is 64 GB, the seek time is 17.5 milliseconds. The seek time corresponding to the seek distance not directly shown in the table entry can be calculated by interpolating between the entries. The seek distance can be expressed not only by the number of bytes but also by the number of cylinders.

(Access time information ACS)
6A and 6B show examples of access time information ACS. The access time information ACS is information indicating the relationship between the data size to be accessed and the time required for the access. Disk access includes data read and data write, and access time information ACS can be defined for each. For example, in the case of data reading, the access time information ACS indicates the correspondence between “read data size” and “read time” required for the read operation. FIG. 6A is a graph showing the correspondence y = g (x). In FIG. 6A, the horizontal axis represents the read data size x, and the vertical axis represents the read time y. Alternatively, as shown in FIG. 6B, the access time information ACS may give the correspondence between the read data size and the read time in a table format. According to FIG. 6B, for example, it can be seen that when the read data size is 4 kB, the read time is 0.10 milliseconds, and when the read data size is 128 kB, the read time is 3.2 milliseconds. The read time corresponding to the read data size not directly shown in the table entry can be calculated by interpolating between the entries.

2. I / O scheduler The I / O scheduler 10 according to the present embodiment performs scheduling processing by appropriately using the above information. The following points are noted in this scheduling process. First, in this embodiment, a predetermined response time (hereinafter referred to as “request period Tprd”) is required for the disk access of the process. All processes are required to make a disk access at least once within the requested period Tprd. The request period Tprd can be set as appropriate by the user as long as the system efficiency does not deteriorate. For example, the request period Tprd is set to be equal to or less than the timeout time of the application executed by the computer system 1. Second, scheduling is performed so that the total throughput of each process regarding disk access is as large as possible. That is, the I / O scheduler 10 according to the present embodiment performs the scheduling process so that the throughput becomes as large as possible under the condition that all processes execute disk access at least once within the request period Tprd.

2-1. Overview As shown in FIG. 7, consider a case where a plurality of processes P <b> 1 to P <b> 4 perform disk access via the I / O scheduler 10. The plurality of processes P1 to P4 access each of the plurality of files F1 to F4 recorded on the disk 4. At this time, each of the processes P 1 to P 4 issues a disk access request to the I / O scheduler 10. The I / O scheduler 10 performs scheduling processing of disk access requests issued from the processes P1 to P4.

  The I / O scheduler 10 has a plurality of wait queues Q1 to Q4 that temporarily hold disk access requests. The wait queue is assigned for each process. In the example of FIG. 7, the wait queues Q1 to Q4 are assigned to the processes P1 to P4, respectively. That is, disk access requests from the processes P1 to P4 are temporarily stored in the waiting queues Q1 to Q4.

  Further, the I / O scheduler 10 includes a process monitor unit 20, a scheduling determination unit 30, and a request selection unit 40. The process monitor unit 20 has a function of monitoring an operating process and acquiring process information. The scheduling determination unit 30 has a function of determining the order of waiting queues from which disk access requests are taken out and the number of disk access requests to be taken out from each waiting queue. The request selection unit 40 has a function of acquiring disk access requests from the waiting queue according to the determined order and number, and sequentially transmitting the acquired ones to the disk 4.

  FIG. 8 is a flowchart showing the scheduling process according to the present embodiment. The scheduling process according to the present embodiment will be described with reference to FIGS.

(Step S10)
The scheduling determination unit 30 reads process characteristic information PROC, file position information LOC, seek time information SEK, and access time information ACS from the storage device 3. In addition, the OS provides the scheduling determination unit 30 with information on the preset request period Tprd. The request period Tprd is common to all the processes P1 to P4.

(Step S20)
The I / O scheduler 10 temporarily stores disk access requests from the processes P1 to P4 in the waiting queues Q1 to Q4, respectively. In addition, the process monitor unit 20 monitors the processes P1 to P4 that are operating, and acquires information about the processes P1 to P4. Information to be acquired includes a process number and a process name. For example, when the OS is Linux, the process number and process name can be acquired by using the ps command.

(Step S30)
The scheduling determination unit 30 determines a “scheduling parameter” indicating a scheduling rule in the request period Tprd. The scheduling parameters include the order of the waiting queues Q1 to Q4 from which disk access requests are taken out, and the number Ai of disk access requests to be taken out from each of the waiting queues Q1 to Q4 in the request period Tprd. Here, the suffix i indicates each of the waiting queues Q1 to Q4, that is, the processes P1 to P4. According to the present embodiment, the scheduling determination unit 30 causes all the processes P1 to P4 to access the disk at least once within the request period Tprd, and the total throughput for each of the processes P1 to P4 is as large as possible. Next, scheduling parameters are determined. Details thereof will be described later.

(Step S40)
The request selection unit 40 selects and outputs disk access requests from the respective waiting queues Q1 to Q4 in accordance with the scheduling parameters determined by the scheduling determination unit 30. That is, the request selection unit 40 selects the waiting queues (i = Q1 to Q4) one by one according to the order indicated by the scheduling parameter, and the number Ai indicated by the scheduling parameter from one waiting queue i being selected. Only take out disk access requests. Then, the request selection unit 40 sequentially sends the extracted disk access requests to the disk 4.

  In response to the disk access request, the disk 4 performs data reading or data writing with respect to the access target file. The result is returned to the process that issued the disk access request. In the present embodiment, all the processes P1 to P4 perform disk access at least once during the request period Tprd.

  Step S40 is repeated for each request period Tprd. When there is no disk access request stored in any of the waiting queues, the process returns to step S20. Further, the process monitor unit 20 described above continues to monitor the process in operation. When the number of operating processes increases or decreases, the process returns to step S20.

2-2. Determination of scheduling parameters (step S30)
Next, step S30 described above will be described in detail. FIG. 9 is a flowchart showing a method for determining a scheduling parameter in the present embodiment.

(Step S310)
First, the scheduling determination unit 30 determines the order of the waiting queue from which the disk access request is taken out in the request period Tprd. In order to increase the throughput in the request period Tprd, it is desirable to reduce as much as possible the time taken for seeking between the files to be accessed in the request period Tprd. For that purpose, it is desirable to make the total of the seek distances between the files to be accessed in the request period Tprd as small as possible. Therefore, the scheduling determination unit 30 determines the order so that the total seek distance between the access target files is as small as possible.

  The scheduling determination unit 30 receives information (for example, process name) of the processes P1 to P4 that are operating from the process monitor unit 20. Then, the scheduling determination unit 30 refers to the process characteristic information PROC to grasp the access target files F1 to F4 accessed by each of the operating processes P1 to P4. Further, the scheduling determination unit 30 refers to the file position information LOC to grasp the arrangement relationship of the access target files F1 to F4 on the disk 4. Then, the scheduling determining unit 30 sets the order of the waiting queues Q1 to Q4 from which the disk access request is taken out so that the access target files F1 to F4 are sequentially accessed in the “arrangement order (arrangement order)” on the disk 4. decide.

  For example, in the example of the file position information LOC shown in FIGS. 4A and 4B, the access target files F1 to F4 are arranged in the order of “F1, F2, F3, F4” from the top of the disk 4. Therefore, as shown in FIG. 10A, when the access target files F1 to F4 are accessed in the order of “F1, F2, F3, F4, F1”, the total seek distance becomes the smallest. However, as shown in FIG. 10B, when the access target files F1 to F4 are accessed in the order of “F1, F4, F2, F3, F1”, the total seek distance becomes large. As described above, in order to increase the throughput, the total seek distance between the access target files F1 to F4 is preferably as small as possible. Therefore, as the access order of the access target files F1 to F4, the order shown in FIG. 10A is preferable. Or it is the same even if it is order "F4, F3, F2, F1, F4" reverse to FIG. 10A. In any case, it is desirable that the access order of the access target files F1 to F4 is in accordance with the “arrangement order (arrangement order)” on the disk 4.

  The scheduling determining unit 30 determines the order of the waiting queues Q1 to Q4 from which the disk access requests are taken out so that the access order as shown in FIG. 10A is reproduced in the request period Tprd. That is, in this example, the extraction order is “Q1, Q2, Q3, Q4” or “Q2, Q3, Q4, Q1” or “Q3, Q4, Q1, Q2” or “Q4, Q3, Q2, Q1”. It is determined. As a result, the time allocated for seeking between the access target files F1 to F4 in the requested period Tprd is suppressed as much as possible, and the time allocatable for data reading or data writing is increased. This means that the total throughput in the requested period Tprd increases.

  For example, as shown in FIG. 10A, the access to the access target files F1 to F4 is completed once in the request period Tprd. In the example of FIG. 10A, in one request period Tprd, the disk head sequentially moves from the position of the file F1 to the position of the file F4, and then returns to the position of the first file F1. That is, access to the access target files F1 to F4 is performed cyclically, and the same operation is repeated for each request period Tprd.

(Step S320)
Next, the scheduling determining unit 30 calculates the seek distance (access seek distance between files) between the access target files F1 to F4 when the disk access requests are extracted in the order determined in step S310. That is, the scheduling determination unit 30 calculates the inter-file seek distance when the access target files F1 to F4 are accessed in the determination order in the request period Tprd. The access target files F1 to F4 of the processes P1 to P4 are indicated in the process characteristic information PROC, and the positions of the access target files F1 to F4 on the disk 4 are indicated in the file position information LOC. Therefore, the seek distance between files can be calculated by referring to the information.

  Since each file has a width corresponding to the file size on the disk 4, it is necessary to determine a representative position of each file when calculating the seek distance between files. Examples of the representative position include a start position, an end position, or an intermediate position between the start position and the end position of each file. For example, when the intermediate position is used as the representative position, the distance between the intermediate position of a certain file and the intermediate position of the next accessed file is calculated as the seek distance between files.

  Consider the case of the example shown in FIG. 10A. In this case, the access order in the request period Tprd is “F1, F2, F3, F4, F1”. Accordingly, the seek distance L12 between the files F1 and F2, the seek distance L23 between the files F2 and F3, the seek distance L34 between the files F3 and F4, and the seek distance L41 between the files F4 and F1 are calculated. .

(Step S330)
Next, the scheduling determination unit 30 refers to the seek time information SEK (see FIGS. 5A and 5B) to calculate an inter-file seek time corresponding to the inter-file seek distance calculated in step S320. When the function y = f (x) as shown in FIG. 5A is used, the corresponding inter-file seek time y can be obtained by substituting the inter-file seek distance into the variable x. When a table as shown in FIG. 5B is used, an inter-file seek time corresponding to an inter-file seek distance can be obtained by interpolating between entries.

  In the example shown in FIG. 10A, the inter-file seek times T12, T23, T34, and T41 corresponding to the inter-file seek distances L12, L23, L34, and L41, respectively, are calculated. The calculated seek time between files corresponds to the time allocated for seeking between the access target files F1 to F4 in the request period Tprd. As a result of the devising of ordering described in step S310, the total inter-file seek time (T12 + T23 + T34 + T41) is minimized. Accordingly, the time allocatable for data reading or data writing in each file increases. This means that the total throughput in the requested period Tprd increases.

(Step S340)
Next, the scheduling determination unit 30 calculates the “remaining time Trst” by subtracting each of the inter-file seek times calculated in step S330 from the request period Tprd. In the case of the example shown in FIG. 10A, the remaining time Trst is calculated as Tprd− (T12 + T23 + T34 + T41). This remaining time Trst corresponds to the time available for data reading or data writing in the access target file. As described above, since the total sum of seek times between files is minimized, the maximum time is secured as the remaining time Trst.

  The processing in step S340 is equivalent to not only estimating the remaining time Trst that can be allocated to data reading or data writing, but also securing in advance the time required for all file-to-file seek operations in the requested period Tprd. To do. Securing the time required for all inter-file seek operations means that all the processes P1 to P4 can perform disk access at least once within the requested period Tprd. In other words, step S340 ensures that all processes P1-P4 make a disk access at least once within the requested period Tprd.

(Step S350)
As described above, the total throughput in the request period Tprd is expected to be improved by the order determination in step S310. Further, the calculation of the remaining time Trst in step S340 ensures that all the processes P1 to P4 perform disk access at least once within the request period Tprd. Thereafter, the scheduling determination unit 30 may appropriately distribute the remaining time Trst among the processes P1 to P4. Based on the time distributed to the respective processes P1 to P4, the number Ai of disk access requests that can be taken out from the respective waiting queues Q1 to Q4 within the remaining time Trst (that is, the request period Tprd) can be determined. . Various methods are conceivable.

<First example>
For example, the number Ai of disk access request acquisitions is determined to be equal among the plurality of processes P1 to P4. For this purpose, the scheduling determination unit 30 performs the following processing, for example. Although the case where the disk access is data reading is illustrated, the same applies to the case of data writing.

  Assume that a data size read from a file to be accessed by a process i by a single disk access request is Dacs_i. This one-time read data size Dacs_i may be given by the user or may be predetermined by the system. Further, the read data size Dacs_i may be the same or different between processes.

  The scheduling determination unit 30 calculates the read time Tacs_i corresponding to one read data size Dacs_i by referring to the access time information ACS (see FIGS. 6A and 6B). When the function y = g (x) as shown in FIG. 6A is used, the corresponding read time Tacs_i can be obtained by substituting the read data size Dacs_i for the variable x. When the table as shown in FIG. 6B is used, the read time Tacs_i corresponding to the read data size Dacs_i can be obtained by interpolating between the entries. Further, the scheduling determining unit 30 refers to the seek time information SEK (see FIGS. 5A and 5B) to calculate the in-file seek time Tsek_i required for moving between read data in the access target file. This in-file seek time Tsek_i is one seek time corresponding to one read data size Dacs_i.

  The sum of the read time Tacs_i and the file seek time Tsek_i (Tacs_i + Tsek_i) corresponds to the time required to process one disk access request from a certain process i. Therefore, for each process i, the number Ai of disk access requests that can be taken out from the waiting queue within the remaining time Trst can be given by the following equation (1).

  In the formula (1), N means the entire processes P1 to P4 in operation. Therefore, the denominator of Expression (1) represents the total time required to process the disk access requests from all the processes P1 to P4 once. By dividing the remaining time Trst by the sum, it is possible to calculate the number of disk access request acquisitions Ai for each process i. In this case, the acquisition number Ai is equal between the processes P1 to P4. Actually, since the acquisition number Ai needs to be an integer, the acquisition number Ai is determined by rounding down the decimal part of the value obtained by the equation (1).

<Second example>
Consider a case where a “weight” is given to the throughput required for each of the processes P1 to P4. In this case, the number Ai of disk access request acquisitions can be determined so that the weighting is reflected in the obtained throughput. The parameter that designates the weighting of the throughput required for each of the processes P1 to P4 is hereinafter referred to as “weighting parameter”. The weighting parameter is included in the process characteristic information PROC, for example.

  FIG. 11 shows the process characteristic information PROC in this example. In FIG. 11, the process characteristic information PROC indicates a weighting parameter in addition to the access target files F1 to F4 related to each of the processes P1 to P4. In this example, the weighting parameter is a “weighting factor” of the required throughput. The weighting factor is a ratio between the throughputs required for the respective processes P1 to P4. For example, when the weighting factors specified for the processes P1 and P2 are 1 and 2, respectively, the process P2 is required to have twice the throughput of the process P1.

  The scheduling determination unit 30 distributes the remaining time Trst according to the weighting factor, and determines the number Ai of disk access request acquisitions so that the weighting factor is reflected in the obtained throughput. Specifically, as in the first example described above, the scheduling determination unit 30 calculates the read time Tacs_i and the in-file seek time Tsek_i per time. Further, the scheduling determination unit 30 refers to the process characteristic information PROC and acquires the weighting factor Wi specified for each of the processes P1 to P4. In this case, the acquisition number Ai of the disk access request is given by the following equation (2).

  Actually, since the acquisition number Ai needs to be an integer, the acquisition number Ai is determined by rounding down the decimal point of the value obtained by the equation (2). When there is a process whose value obtained by the expression (2) is less than 1, the number of acquisitions related to that process is set to 1, and instead, 1 is subtracted from two or more acquisition numbers related to other processes. When the process with the maximum number of acquisitions is selected as the subtraction target, the reduction rate of the throughput allocated to the process is small, so that the influence can be suppressed small. If the number of acquisitions related to all processes does not become 1 or more by such processing, the I / O scheduler 10 can also notify the administrator of an error.

<Third example>
The weighting of the required throughput is not limited to the weighting factor. For example, let us consider a case where a “priority process” in which the requested throughput is given preferentially and a “best effort process” in which the requested throughput is given with the best effort are mixed in the processes P1 to P4. In this case, the classification between the priority process and the best effort process corresponds to the weighting of the requested throughput.

  FIG. 12 shows the process characteristic information PROC in this example. In FIG. 12, the process characteristic information PROC indicates a weighting parameter in addition to the access target files F1 to F4 related to the processes P1 to P4, respectively. In this example, the weighting parameters include a “class” indicating whether each process is a priority process or a best effort process, a “request throughput” regarding the priority process, and a “weighting factor” of the request throughput regarding the best effort process. . The requested throughput related to the priority process is a throughput that should be secured with priority. The weighting factor is the ratio between the throughput required for each of the best effort processes.

  The scheduling determination unit 30 distributes the remaining time Trst according to the weighting parameter, and determines the number Ai of disk access request acquisitions so that the weighting parameter is reflected in the obtained throughput. With reference to FIG. 13, the process of step S350 in this example will be described in detail.

(Step S351)
The scheduling determining unit 30 refers to the process characteristic information PROC, and first determines the disk access request acquisition number Ak for the priority process k. Here, the subscript k indicates a priority process included in the processes P1 to P4. In the example of FIG. 12, processes P1 and P2 are priority processes k. In the process characteristic information PROC, it is assumed that the requested throughput specified for the priority process k is Bk. Further, it is assumed that the data size read from the access target file by the priority process k by one disk access request is Dacs. This one-time read data size Dacs may be given by the user or may be predetermined by the system. In this case, the scheduling determination unit 30 calculates the acquisition number Ak for the priority process k according to the following equation (3).

  In equation (3), the numerator represents the amount of data required by the priority process k within the required period Tprd. By dividing this required data amount by the read data size Dacs for one time, the number of acquisitions Ak necessary for realizing the required throughput Bk can be calculated. That is, the scheduling determination unit 30 determines the acquisition number Ak from each waiting queue so that the requested throughput Bk can be obtained for the priority process k. Actually, since the acquisition number Ak needs to be an integer, the acquisition number Ak is determined by rounding down the decimal point of the value obtained by the equation (3).

(Step S352)
Next, the scheduling determination unit 30 calculates the time required for processing the disk access request for the acquired number Ak based on the acquired number Ak determined in step S351. Then, the scheduling determination unit 30 calculates a new remaining time Trst ′ by subtracting the calculated time from the remaining time Trst. Specifically, the new remaining time Trst ′ is given by the following equation (4).

  In Equation (4), Tacs_k and Tsek_k are a read time and a seek time within a file for the priority process k, respectively. The read time Tacs_k and in-file seek time Tsek_k per one time can be calculated by referring to the access time information ACS and seek time information SEK, as in the first example.

  By the process shown in the equation (4), the time necessary for processing the disk access request for the number of acquired Ak is secured in advance. That is, the time necessary for realizing the required throughput Bk of the priority process k in the required period Tprd is secured in advance. It can be said that the scheduling determination unit 30 updates the remaining time Trst by securing the time for the priority process k in advance. The remaining time Trst 'after the update is a time that can be allocated to the best effort process.

(Step S353)
Thereafter, the scheduling determination unit 30 may distribute the remaining time Trst ′ after the update among the best effort processes j with reference to the process characteristic information PROC. Here, the subscript j indicates the best effort process included in the processes P1 to P4. In the example of FIG. 12, the processes P3 and P4 are the best effort process j. Based on the time distributed to the best effort processes P3 and P4, the number Aj of disk access requests that can be taken out from the respective waiting queues Q3 and Q4 within the remaining time Trst ′ can be determined.

  For example, as shown in FIG. 12, consider a case where a weighting factor for requested throughput is specified for the best effort process. In this case, the scheduling determination unit 30 distributes the remaining time Trst ′ according to the weighting factor, and determines the acquisition number Aj so that the weighting factor is reflected in the obtained throughput. Specifically, as in the first example described above, the scheduling determination unit 30 calculates a read time Tacs_j and an in-file seek time Tsek_j per time. In addition, the scheduling determination unit 30 refers to the process characteristic information PROC and acquires the weighting factor Wj specified for the best effort process j. In this case, the acquisition number Aj regarding the best effort process j is given by the following equation (5).

  Equation (5) corresponds to a case where the remaining time Trst in the foregoing equation (2) is replaced with the updated remaining time Trst 'and the target is limited to only the best effort process j. Actually, since the acquisition number Aj needs to be an integer, the acquisition number Aj is determined by rounding down the decimal point of the value obtained by the equation (5).

  As described above, the scheduling determination unit 30 separately distributes the remaining time Trst to each of the priority process and the best effort process, and calculates the acquisition numbers Ak and Aj separately. According to the acquisition numbers Ak and Aj, the throughput required for the priority process k is realized, and the ratio of the requested throughput specified for the best effort process j is realized. That is, the acquisition numbers Ak and Aj are determined so that weighting is reflected in the throughput obtained for each process.

  As described above, in step S350, the scheduling determination unit 30 determines the number of acquired disk access requests Ai for each process P1 to P4 by appropriately distributing the remaining time Trst among the processes P1 to P4. can do.

  The “order” determined in step S310 and the “number of acquisitions Ai” determined in step S350 become “scheduling parameters” indicating the scheduling rule in the request period Tprd. The scheduling determination unit 30 determines scheduling parameters so that all processes P1 to P4 perform disk access at least once within the requested period Tprd, and the total throughput for each of the processes P1 to P4 is as large as possible. Is done. When the weighting parameter is given, the scheduling determination unit 30 can determine the scheduling parameter so that the weight is reflected in the throughput regarding each of the processes P1 to P4.

  According to the determined scheduling parameters, all processes can perform disk access at least once within a given request period Tprd, and throughput is maximized under given conditions. That is, it is possible to improve throughput while preventing occurrence of an application timeout or the like.

  The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the above-described embodiments, and can be appropriately changed by those skilled in the art without departing from the scope of the invention.

FIG. 1 is a conceptual diagram of an I / O scheduler for explaining the problem of the present invention. FIG. 2 is a block diagram showing a configuration example of the computer system according to the embodiment of the present invention. FIG. 3 is a table showing an example of process characteristic information in the present embodiment. FIG. 4A is a table showing an example of file position information in the present embodiment. FIG. 4B is a table showing another example of file position information in the present embodiment. FIG. 5A is a graph showing an example of seek time information in the present embodiment. FIG. 5B is a table showing another example of seek time information in the present embodiment. FIG. 6A is a graph showing an example of access time information in the present embodiment. FIG. 6B is a table showing another example of access time information in the present embodiment. FIG. 7 is a functional block diagram conceptually showing the functions of the I / O scheduler according to the present embodiment. FIG. 8 is a flowchart showing the scheduling process according to the present embodiment. FIG. 9 is a flowchart showing a method for determining a scheduling parameter in the present embodiment. FIG. 10A is a conceptual diagram illustrating an example of an access order for an access target file. FIG. 10B is a conceptual diagram illustrating another example of an access order for an access target file. FIG. 11 is a table showing another example of process characteristic information in the present embodiment. FIG. 12 is a table showing still another example of the process characteristic information in the present embodiment. FIG. 13 is a flowchart illustrating an example of a method for determining the number of acquired disk access requests according to the present embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Computer system 2 Processing apparatus 3 Storage apparatus 4 Disk 5 Input apparatus 6 Output apparatus 10 I / O scheduler 20 Process monitor part 30 Scheduling determination part 40 Request selection part PROC Process characteristic information LOC File position information SEK Seek time information ACS Access time information Ai Scheduling parameter Tprd Request period Trst Remaining time

Claims (10)

An I / O scheduler that causes a computer to execute scheduling processing of a disk access request issued from each of the plurality of processes when a plurality of processes access the disk,
The scheduling process includes
(A) reading process characteristic information indicating an access target file accessed by each of the plurality of processes and file position information indicating a position of a file recorded on the disk from a storage device;
(B) storing disk access requests from the plurality of processes in each of a plurality of waiting queues;
(C) Referring to the process characteristic information and the file position information, the order of the plurality of wait queues from which disk access requests are taken out so that the access target file is sequentially accessed in the arrangement order on the disk. A step to determine;
(D) determining the number of disk access requests to be retrieved from each of the plurality of wait queues within the predetermined time period so that all of the plurality of processes perform disk access at least once within a predetermined time period; When,
(E) obtaining a disk access request from the plurality of waiting queues according to the determined order and number, and sending the obtained disk access request to the disk. The I / O scheduler according to claim 1,
The step (A) further includes a step of reading seek time information indicating a correspondence relationship between a seek distance on the disk and a seek time from the storage device,
The step (D)
(D1) referring to the file position information, and calculating a seek distance between the access target files when the access target file is accessed in the determined order;
(D2) referring to the seek time information, calculating a seek time corresponding to the calculated seek distance;
(D3) calculating a remaining time by subtracting the calculated seek time from the predetermined time;
(D4) determining the number of disk access requests to be retrieved from each of the plurality of wait queues within the remaining time by distributing the remaining time among the plurality of processes. The I / O scheduler according to claim 2, wherein
In the step (D4), the number of disk access requests is determined to be equal among the plurality of processes. I / O scheduler. The I / O scheduler according to claim 2, wherein
The process characteristic information further indicates a weighting parameter that specifies a weight of throughput required for each of the plurality of processes,
In the step (D4), the number of disk access requests is determined so that the weighting is reflected in the throughput related to each of the plurality of processes. The I / O scheduler according to claim 4, wherein
The weighting parameter includes a ratio between throughput required for each of the plurality of processes;
In the step (D4), the number of disk access requests is determined so that the ratio is reflected in the throughput related to each of the plurality of processes. The I / O scheduler according to claim 4, wherein
The plurality of processes include a priority process in which a required throughput is preferentially given, and a best effort process in which a required throughput is given at best effort,
The weighting parameter includes a class indicating whether each of the plurality of processes is the priority process or the best effort process;
The step (D4) includes
(D41) determining the number of disk access requests to be retrieved from the corresponding one of the plurality of waiting queues so that the required throughput is obtained with respect to the priority process;
(D42) updating the remaining time by subtracting the time required to process the determined number of disk access requests from the remaining time;
(D43) determining the number of disk access requests to be retrieved from the corresponding one of the plurality of waiting queues for the best effort process by distributing the updated remaining time among the best effort processes; I / O scheduler including The I / O scheduler according to claim 6, wherein
The weighting parameter further includes a ratio between throughput required for each of the best effort processes;
In the step (D43), the number of disk access requests is determined such that the ratio is reflected in the throughput for each of the best effort processes. The I / O scheduler according to claim 1,
The process characteristic information further indicates a weighting parameter that specifies a weight of throughput required for each of the plurality of processes,
In the step (D), the predetermined process is performed so that all of the plurality of processes perform disk access at least once within the predetermined time period, and the weight is reflected in the throughput relating to each of the plurality of processes. The number of disk access requests to be fetched from each of the plurality of waiting queues within a predetermined time is determined. I / O scheduler. An I / O scheduling method for performing a scheduling process of a disk access request issued from each of the plurality of processes when a plurality of processes access the disk,
(A) reading process characteristic information indicating an access target file accessed by each of the plurality of processes and file position information indicating a position of a file recorded on the disk from a storage device;
(B) storing disk access requests from the plurality of processes in each of a plurality of waiting queues;
(C) Referring to the process characteristic information and the file position information, the order of the plurality of wait queues from which disk access requests are taken out so that the access target file is sequentially accessed in the arrangement order on the disk. A step to determine;
(D) determining the number of disk access requests to be retrieved from each of the plurality of wait queues within the predetermined time period so that all of the plurality of processes perform disk access at least once within a predetermined time period; When,
(E) acquiring disk access requests from the plurality of waiting queues according to the determined order and number, and sending the acquired disk access requests to the disk. An I / O scheduling method. A disc,
A processing device for executing a plurality of processes for accessing the disk;
An I / O scheduler that is executed by the processing device and performs scheduling processing of a disk access request issued from each of the plurality of processes;
A storage device for storing process characteristic information indicating an access target file to be accessed by each of the plurality of processes, and file position information indicating a position of a file recorded on the disk, and
The I / O scheduler is
Store disk access requests from the plurality of processes in a plurality of wait queues,
Referring to the process characteristic information and the file position information, determine the order of the plurality of wait queues from which disk access requests are taken out so that the access target file is accessed in order in the arrangement order on the disk;
Determining the number of disk access requests to be retrieved from each of the plurality of wait queues within the predetermined time period so that all of the plurality of processes perform disk access at least once within a predetermined time period;
A computer system that acquires a disk access request from the plurality of waiting queues according to the determined order and number, and sends the acquired disk access request to the disk.

Images