JP2008146420A - Computer program for optimizing processing sequence of i/o request - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the throughput of an I/O control system for optimizing an I/O request processing sequence in the I/O control system. <P>SOLUTION: This computer program includes checking how many and how sequenced I/O requests are lined up in the queue of an I/O control system, and which storage region is pertinent to the I/O destination of each of the I/O requests, and calculating the number of I/O as the number of the I/O requests as the I/O destinations for each of two or more storage regions as the I/O destinations of a plurality of I/O requests lined up in the queue among a plurality of storage regions based on the check result, and rearranging the sequence of the plurality of I/O requests in the queue so that the I/O requests whose I/O destinations belong to the storage regions with the larger number of I/O can be sequenced more earlier. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to optimization of the processing order of I / O requests (input / output requests).

  For example, Patent Document 1 discloses that access requests from applications are centrally monitored and expected load data when an application is added is output. Further, Patent Document 2 discloses that when accesses are concentrated locally, scheduling is performed in consideration of seek / search time by temporarily accumulating in a buffer capable of high-speed writing. Patent Document 3 discloses that the order of issuing I / O requests is determined using an evaluation formula [registration time * schedule parameter]. Further, Patent Document 4 discloses that a DBMS access plan is generated to optimize transaction effective efficiency.

JP 2005-149283 A Japanese Patent Laid-Open No. 10-111762 Japanese Patent Laid-Open No. 10-289189 JP 2002-32249 A

  By the way, when an I / O request is received from an I / O request issuer, is temporarily stored in a queue, and predetermined processing is completed for an I / O request output from the queue, a predetermined response is issued. As an I / O control system to be returned to the original, for example, a DBMS (DataBase Management System) which is software for managing a database is known. The DBMS performs an operation related to a database (hereinafter, “database operation”) based on an I / O request from an application. Database operations include, for example, writing (updating) data, reading (referring) data, and deleting data. When a predetermined process is completed for the I / O request, the DBMS returns a predetermined response to the application that issued the I / O request. As the predetermined processing, for example, there is processing for obtaining an exclusive control right described later and issuing an I / O command to the storage device as one of database operations.

  The DBMS is generally configured as a process group including a plurality of processes that are operated on an operating system (OS) and recognized by the OS. Hereinafter, each process in the DBMS is referred to as a “DBMS process”. Each of the plurality of DBMS processes communicates with a plurality of application processes (processes in the application program), and processes I / O requests from the application processes. For this reason, a plurality of DBMS processes may simultaneously access the same storage area.

  In order to prevent this, exclusive control using a mutex is used. In this case, only a DBMS process that has acquired an exclusive control right (in other words, an I / O execution right) for a certain storage area can execute I / O for the storage area, while in the same storage area. Another DBMS process that wishes to execute I / O waits for execution of I / O to the storage area until it can acquire the exclusive control right corresponding to the storage area.

  There may be cases where I / O is generated for each of a plurality of storage areas with one I / O request. In this case, when the exclusive control right can be acquired for all of the plurality of storage areas, I / O can be performed on each of the storage areas. Completion of processing for such an I / O request, in other words, the waiting time for an application from issuing an I / O request to receiving a predetermined response from the DBMS is the number of storage areas. Depends on the process that ends most recently.

  Normally, I / O requests issued from multiple application processes are temporarily stored in a queue managed by the DBMS (typically FIFO (First In First Out) buffer) and output from the queue. However, if the processing is performed in the order in which the I / O requests are received in such an environment, the throughput of the DBMS may be deteriorated.

  The problems as described above can be present not only in the DBMS but also in other types of I / O control systems described above.

  Therefore, an object of the present invention is to improve the throughput of the I / O control system by optimizing the I / O request processing order in the I / O control system.

  It is checked which number of I / O requests are arranged in what order in the queue of the I / O control system and which storage area is the I / O destination of each I / O request. Then, based on the result of the check, I / O is performed for each of two or more storage areas that are set as I / O destinations in a plurality of I / O requests arranged in the queue among the plurality of storage areas. The number of I / Os, which is the number of I / O requests that are the destination, is calculated. The order of the I / O requests is rearranged.

  The number of I / Os is calculated for each of two or more storage areas that are the I / O destinations of a plurality of I / O requests that remain accumulated in the queue, and attention is paid to the calculated number of I / Os. Rearrangement, specifically, rearrangement is performed so that I / O requests that have a storage area with a large number of I / Os as the I / O destination are processed first. Increased throughput can be expected.

  This technique can be applied to an I / O control system. The “I / O control system” mentioned here may be a computer program typified by a DBMS, and exhibits an I / O control function (for example, a function as a DBMS) by reading and executing the computer program. It may be a computer machine (for example, a server machine).

  “I / O” is an abbreviation for input / output. Therefore, the I / O request may be an input request or a write request. Further, the I / O request is not limited to a request for direct input or output, but may be a request for input or output. For example, when the I / O control system is a DBMS and the DBMS is a DBMS of a type that receives a prepare (SQL statement) and receives a commit after receiving a preparation (SQL statement) in response thereto, prepare. And at least one of the commits is the I / O request described above.

  The check can be performed regularly or irregularly. The rearrangement of the order of I / O requests is typically performed each time a check is performed, but is not limited to this, and the rearrangement is performed once after a plurality of checks are performed. It may be a method.

  <First embodiment>

  Hereinafter, a method for optimizing the processing order of I / O requests according to the first embodiment of the present invention will be described with reference to the drawings. In the following description, it is assumed that the I / O control system is a DBMS (particularly, a relational database management system (RDBMS)). In addition, hereinafter, for the sake of easy understanding, it is assumed that one I / O request is one transaction, one DBMS process that is transmitted from one application process and assigned CPU time from the operating system (OS). Is read from the queue and processed.

  FIG. 1 is a diagram showing a configuration of the entire system according to the present embodiment.

  A server 104 and a client 101 (here, client (1) 101a, client (2) 101b, client (3) 101c) are connected to the network 103, respectively.

  The server 104 functions as a database server provided with the DBMS 107. One process (application process) in the application 102 (here, application (1) 102a, application (2) 102b, application (3) 102c) of each client 101 sends the contents of the database operation to the DBMS 107, for example, SQL. The request described below (hereinafter referred to as “I / O request”) is transmitted. Receiving the I / O request, the DBMS 107 temporarily stores the received I / O request in the waiting queue 105 (here, a FIFO (First-In First-Out) type buffer). Of the process group 1071 of the DBMS 107, one process (DBMS process) to which the CPU time is allocated from the OS reads and interprets the I / O request at the head of the waiting queue 105, and converts it to the I / O request. Perform the corresponding database operation. By this database operation, an I / O command is issued to the storage 108. Here, the storage 108 can be, for example, an apparatus including one or more disk storage devices (hereinafter simply referred to as “disks”) 1081. In the storage 108, a plurality of disks 1081 may form a RAID (Redundant Array of Independent (or Inexpensive) Disks), or a logical unit (logical storage device) using the storage space of one or more disks 1081. ) May be prepared. Instead of the disk 1081 or the logical unit, another type of secondary storage device may be employed (the primary storage device is a memory (not shown) in the server 104). That is, the secondary storage device may be a physical storage device or a logical storage device.

  The DBMS 107 manages a plurality of logical storage areas and information indicating the correspondence between each of the plurality of logical storage areas and a physical storage area in the storage 108. Here, the logical storage area is the block 1074, and the information representing the correspondence between the plurality of blocks 1074 and the plurality of physical storage areas is the block index 1073. In the I / O request, for example, information indicating a physical storage area (hereinafter, physical area information, for example, an address) is recorded as information indicating the I / O destination (hereinafter, I / O destination information). The DBMS 107 identifies the block 1074 corresponding to the physical storage area represented by the physical area information (that is, the block 1074 that is the I / O destination) by referring to the block index 1073 using the physical area information. can do. One physical storage area associated with one block 1074 may be defined in any way in the block index 1073. Specifically, the storage 108 is separated into several physical storage areas, and a database operation is performed for each of the separated storage areas. This separated storage area is a unit of exclusive control, and this area is hereinafter referred to as an exclusive control unit. The exclusive control unit may take various forms, and may be, for example, at least one of the disk 1081, the file 1082, the record 1083, and the logical unit (in this case, for example, a form that spans a plurality of disks 1081). Or may be defined as part of the disk 1081). Since one block 1074 is associated with such an exclusive control unit, the block 1074 is also an exclusive control unit.

In this embodiment, for example, one I / O request (hereinafter referred to as “target I / O request” in this description) is completed in the following flow.
(Step 1) The application process issues a prepare statement (SQL).
(Step 2) The DBMS 107 interprets the SQL from the application process, converts it into a target I / O request, and temporarily stores it in the standby queue 105.
(Step 3) The application process issues a commit statement (SQL).
(Step 4) The commit flag of the target I / O request accumulated in the waiting queue 105 is validated.
(Step 5) Every time an I / O request in the front of the waiting queue 105 with the commit flag being valid is taken out by the DBMS process, the target I / O request accumulated behind it is retrieved. The order goes up by one. By repeating such processing, the possibility that the target I / O request with the commit flag being valid is extracted is increased. Eventually, the target I / O request is extracted from the standby queue 105 by the DBMS process. The DBMS process uses the information included in the target I / O request and the block index 1073 to identify an I / O destination block 1074 (hereinafter referred to as the target block 1074), and a mutex 1072 corresponding to the target block 1074. Try to get. If it cannot be obtained, it will be waiting. If there are a plurality of target blocks 1074, the process waits until all of the plurality of mutexes 1072 respectively corresponding to the plurality of target blocks 1074 can be acquired.
(Step 6) If the mutex 1072 can be acquired, the DBMS process designates the physical storage area for the secondary storage device having the physical storage area corresponding to the target block 1074 in the storage 108. Sent the I / O command. When a predetermined command response (for example, I / O success) is received in response to the I / O command, the DBMS process releases the mutex 1072, and receives a predetermined request response (for example, success) for the target I / O request. Notify the application process.
(Step 7) The application process receives success.

  In the above flow, steps (step 1) to (step 7) are waiting times for the application program 102. The command response described above is a response to the I / O command that the DBMS receives from the secondary storage device that is the transmission destination of the I / O command. On the other hand, the above-described request response is a response to the I / O request that the DBMS transmits to the application that is the transmission source of the I / O request.

  In the present embodiment, an optimization unit 106 is provided in order to reduce the overall waiting time for a plurality of I / O requests accumulated in the standby queue 105. The optimization unit 106 is specified from a plurality of I / O requests arranged in the standby queue 105 (typically, a FIFO (First-In First-Out) buffer) and the plurality of I / O requests. By creating a matrix based on a plurality of blocks and performing rearrangement on the matrix, the overall waiting time can be shortened. The optimization unit 106 is a computer program that is executed by the CPU (not shown) of the server 104, like the DBMS 107. The optimization unit 106 may be incorporated in the DBMS 107 or may exist outside the DBMS 107 so as to be communicable with the DBMS 107.

  FIG. 2 is an explanatory diagram of the function of the optimization unit 106.

  The optimization unit 106 includes a queue monitoring storage unit 1061, an I / O performance storage unit 1062, and an I / O matrix replacement unit 1063.

  The queue monitoring accumulation unit 1061 periodically checks the status of the standby queue 105, and stores information (hereinafter referred to as “queue status information”) 2611 indicating the status on a predetermined storage area 261 (for example, on a memory (not shown) in the server 104). Area). The queue status information 2611 includes, for example, the number and order of I / O requests stored in the standby queue 105, the arrival time of each I / O request, a commit flag, and request contents. The request content includes information such as the type of database operation such as writing or reading, the I / O destination block 1074, and the like.

  The I / O performance accumulating unit 1062 acquires the I / O performance corresponding to the secondary storage device for each secondary storage device. The I / O performance referred to here is, for example, the length of time from when an I / O command is issued to the secondary storage device by the DBMS process until a predetermined response is returned from the secondary storage device. The I / O performance storage unit 1062 updates the I / O performance information 2612 corresponding to the secondary storage device of the I / O performance, using the acquired numerical value (I / O performance value) representing the I / O performance. To do. The I / O performance information 2612 is prepared for each secondary storage device in a predetermined storage area 261, for example. The I / O performance information 2612 includes an I / O performance value acquired at each time point, a plurality of I / O performance value statistics (I / O performance statistics) from the first time point to the second time point, and Is included. The I / O performance statistics are obtained by the I / O performance accumulation unit 1062, for example. I / O performance statistics can typically be considered to follow a standard normal distribution. This I / O performance statistic is useful, for example, for an administrator to determine whether each block size is appropriate in a test operation before starting business in the system according to the first embodiment. For example, the administrator refers to the I / O performance statistics of the same type of secondary storage device (for example, a secondary storage device having the same data transfer rate and I / O rate), thereby allowing the same type of secondary storage device. It is determined whether the I / O performance (for example, average, standard deviation) has a difference greater than the expected error despite being a device, and if there is such a difference, the I / O performance is low. The block size associated with the secondary storage device is made smaller than the block size associated with the secondary storage device with higher I / O performance (in other words, the physical storage area allocated to one block Tuning, such as reducing the size). The above determination and tuning may be automatically performed by a computer program (for example, a program included in the optimization unit 106).

  Based on the queue status information accumulated by the queue monitoring accumulation unit 1061, the I / O matrix substitution unit 1063 rearranges the order of I / O requests arranged in the standby queue 105, for example, as illustrated.

  Details of the rearrangement process will be described below with reference to FIGS. 3 and 4. In each figure, “R” means an acronym for Request (request), and “B” means an acronym for Block (block). In the following, for the sake of simplicity, the standby queue 105 stores seven I / O requests R1 to R7 as snapshots at a certain point in time, and all the I / O requests R1 to R7 have a commit flag. Each I / O request is valid and will be described on the assumption that I / O is generated in the range of blocks B1 to B7.

  For example, the I / O matrix replacement unit 1063 interprets the queue status information and creates a matrix as shown in FIG. 3A based on the result of the interpretation. This matrix arranges the I / O requests R1 to R7 stored in the standby queue 105 in the row direction (that is, arranges the I / O requests R1 to R7 according to the order specified from the queue status information), Blocks B1 to B7 are arranged in the column direction. In each row, a predetermined value (represented by a circle in the figure) is set at the position corresponding to the block that is the I / O destination in the I / O request corresponding to that row. Therefore, it can be seen from this matrix that, for example, in the I / O request R1, the blocks B1 and B5 are the I / O destinations.

  After creating such a matrix, column replacement processing and row replacement processing are performed.

  The column replacement process is as follows. That is, the I / O matrix replacement unit 1063 calculates the number of I / Os (number of circles) for each block. Then, as shown in FIG. 3B, the I / O matrix replacement unit 1063 moves the block (column) with the larger number of I / Os to the left. As a result, they are rearranged in order from the left to B1, B6, B2, B5, B4, B7, and B3.

  In the row replacement processing, attention is paid to blocks having a large number of I / Os, and rearrangement is performed so that the order of I / O requests (rows) with the focused block as an I / O destination is as far as possible. At this time, the arrangement (uploading) focused on the block having a larger number of I / Os than the focused block is not broken, and if it is broken, the row replacement process is finished.

  Specifically, for example, as shown in FIG. 4A, first, focusing on the block B1 having the largest number of I / Os, the I / O request in which the block B1 is the I / O destination is higher. Rearrange so that it comes to (so that the output order comes first). Here, since only the I / O request R2 does not perform I / O to the block B1, the row corresponding to the I / O request R2 is moved to the lowest order.

  Next, as shown in FIG. 4B, paying attention to the block B6 having the second largest number of I / Os, the block B6 becomes the I / O destination so as not to disturb the top-filled state of B1. Rearrange the I / O requests (rows) that are currently in the higher rank.

  Next, as shown in FIG. 4C, paying attention to the block B2 having the third largest number of I / Os, the block B2 is connected to the I / O destination in such a way that the top-packed state of B1 and B6 is not destroyed. Rearrange the I / O requests (rows) at the top.

  Next, paying attention to the block B5 having the fourth largest number of I / Os, an I / O request in which the block B5 is an I / O destination without disturbing the top-filled state of B1, B6, and B2 Try rearranging so that (row) comes to the top. However, here, as can be seen from the matrix in FIG. 4D, which is the top-padded matrix focusing on the block B2, any of the I / O requests R4, R1, and R7 with the block B5 as the I / O destination is selected. If it is going to be higher, the top-packed state of any of B1, B6 and B2 will be destroyed.

  Therefore, at this stage, the row replacement process is terminated.

  Through the series of processes described above, the optimization of the processing order of I / O requests is completed. The effects that can be expected from this optimization will be verified below. At that time, the probability (P) of delay in processing related to I / O for one block (specifically, I / O for a secondary storage device having a physical storage area corresponding to the block) is uniformly 0.05. And The probability of delay (P) is, for example, that the standard deviation (σ) is out of the range of ± 2σ in the standard normal distribution of I / O performance of the secondary storage device (standard normal distribution in the I / O performance information). It is a probability.

(1) Before rearrangement [see FIG. 3 (a)]
-Block B1: Since I / O requests R3 to R7 and I / O are continuously generated,
Probability of delay: P _B1 = 1− (1−0.05) ⁵ = 0.2262190625
Block B5: I / O is generated in the final I / O request R7, but I / O is not generated in block B5 in the previous I / O request R6.
Probability of delay: P _B5 = 0.05
Block B7: I / O occurs in the final I / O request R7, but no I / O occurs in block B7 in the previous I / O request R6.
Probability of delay: P _B7 = 0.05
For other blocks, no new I / O occurs at the timing when the final I / O request (R7) is sent. The earlier delay is absorbed in the R7 time slot.

Each delay may be considered to occur independently. Therefore, the probability that the delay occurs simultaneously in two or more of B1, B5, and B7 is a product of the respective probabilities, and thus becomes a small value that can be ignored. Therefore, the probability P that the delay occurs at the timing of the final I / O request R7 is
P ≒ P _B1 + P _B5 + P _B7 = 0.226 + 0.05 + 0.05 = 0.326
It becomes.

(2) After rearrangement [see FIG. 4 (d)]
Block B6: I / O occurs in the final I / O request R2, but no I / O occurs in block B6 in the previous I / O request R7.
P _B6 = 0.05
Block B2: I / O is generated in the final I / O request R2, but I / O is not generated for the block B2 in the previous I / O request R7.
P _B2 = 0.05
Each delay may be considered to occur independently. Therefore, the probability that the delay occurs simultaneously in both B6 and B2 is the product of the respective probabilities, but they are small enough to be ignored. Therefore, the probability P of delay in the final I / O request R2 is
P ≒ P _B6 + P _B2 = 0.05 + 0.05 = 0.1
It becomes.

  As described above, according to the first embodiment, the number of I / Os is calculated for each of two or more blocks that are set as I / O destinations in a plurality of I / O requests that remain accumulated in the standby queue 105. Thus, rearrangement is performed such that an I / O request in which a block having a large number of I / Os is an I / O destination is processed first. Thereby, improvement of the throughput of the DBMS 107 can be expected.

  <Second Embodiment>

  The second embodiment will be described below. At that time, differences from the first embodiment will be mainly described, and description of common points with the first embodiment will be omitted or simplified.

  The I / O matrix replacement unit 1063 executes row composition processing. Hereinafter, this will be described with reference to FIG. Based on the matrix after row replacement processing (see FIG. 4D), the I / O matrix replacement unit 1063 determines whether there are two or more rows in which the inner product of the row vectors is 0 (I / O It is determined whether or not there is an I / O request in which the previous block does not overlap. If there are two or more rows, the I / O matrix replacement unit 1063 combines two or more rows, that is, combines two or more I / O requests (eg, R2 and R7) into one I / O. Summarize in O request. Thereby, further improvement of the throughput of the DBMS 107 can be expected.

  When two I / O requests are combined into one as described above, if the two I / O requests are I / O requests issued from two applications 102, they are combined into one. When the processing of the I / O request is completed, the DBMS 107 notifies the completion to the two applications 102.

  For example, when the number of blocks is enormous and it is difficult to perform row replacement processing in one matrix, the I / O matrix replacement unit 1063 prepares the matrix in multiple stages, At least one of the above, column replacement processing, row replacement processing, and row composition processing may be performed. For example, as illustrated in FIG. 5B, N positions corresponding to N (N is an integer of 2 or more) blocks in the first-stage matrix are assigned to the second stage (one of the first stages). It may be associated with one position in the upper matrix. In that case, in the second-stage matrix, a numerical value is set at each position where the matrix intersects when the number of N blocks corresponding to the position is all I / O destinations. In the example of FIG. 5B, a value of 0.5 is set at a certain position of the second-stage matrix because four of the eight blocks are I / O destinations. Is done. In this second-stage matrix, column replacement processing, row replacement processing, and row composition processing may be performed. In the row composition processing, for example, the I / O matrix replacement unit 1063 determines whether there are two or more rows in which the inner product of the row vectors is equal to or less than a predetermined value (for example, the row composition processing is performed). It is determined whether there are two or more rows that are very close to 0 with the possibility of overlapping I / O destinations even if they are executed). If there are two or more such rows, the rows may be combined.

  <Third embodiment>

  In this third embodiment, I / O performance statistics are utilized more effectively. Specific examples include the following first to third examples.

  The first example is an improvement in I / O performance. For example, the I / O performance accumulation unit 1062 executes the process illustrated in FIG. 6A periodically or irregularly. Specifically, for example, the I / O performance accumulation unit 1062 refers to the I / O performance statistics of the same type of secondary storage device, so that the I / O performance (for example, the average , Standard deviation) is determined whether or not there is a difference greater than the expected error (step S101). If there is such a difference, the I / O performance storage unit 1062 causes the secondary storage device with the lower I / O performance among the same type of secondary storage devices to execute the optimization process ( S102).

  Here, which of the plurality of secondary storage devices is the same type of secondary storage device is identified by a method such as referring to a management table in which attribute information related to the secondary storage device is written. be able to.

  The optimization process executed on the secondary storage device is, for example, a defragmentation process, that is, a process for eliminating fragmentation of data in the secondary storage device. For example, the I / O performance storage unit 1062 can execute optimization processing for a desired secondary storage device by calling a utility function supported by the OS 211.

  The second example is the determination of the output timing of the I / O request from the standby queue 105. In other words, in the third embodiment, the I / O request is output from the standby queue 105 not at the timing determined by the OS 211 but at the timing determined by the queue control unit (not shown) of the optimization unit 106. . A specific example is shown in FIG.

  That is, when the queue control unit reads an I / O request, the queue control unit displays I / O performance statistics corresponding to the secondary storage device to which the I / O command issued at the time of processing the I / O request. Based on this, the I / O performance for the secondary storage device is predicted (S201). The I / O performance predicted here may be, for example, an average in a standard normal distribution or a certain I / O performance within a predetermined range in the standard normal distribution.

  Based on the predicted I / O performance, the queue control unit determines the timing for reading the next I / O request (for example, how many seconds after the current reading) (S202). Then, the queue control unit reads the next I / O request at that timing (S203), and executes S201 again for the read I / O request.

  The third example is determination of what block is preferentially rearranged in the row replacement processing by the I / O matrix replacement unit 1063. That is, in the first embodiment described above, the order in which the number of I / Os is large is the order in which the block priority is high, but in this third example, the order in which the number of I / Os is large is not necessarily the order. The order of priority is not high. For example, when the number of I / Os in block B1 is the largest, the I / O performance of the secondary storage device corresponding to block B2 is considerably lower than the I / O performance of the secondary storage device corresponding to block B1. The row replacement process with the block B2 as the highest priority may be performed.

  The several embodiments of the present invention described above are examples for explaining the present invention, and are not intended to limit the scope of the present invention only to those embodiments. The present invention can be implemented in various other modes without departing from the gist thereof.

1 shows a configuration of an entire system according to an embodiment of the present invention. Explanatory drawing of the function of an optimization part. Explanatory drawing of the replacement process of the column in a rearrangement process. Explanatory drawing of the replacement process of the line in a line-up purchase process. (A) is explanatory drawing of the synthetic | combination process of the line in a rearrangement process. (B) is an explanatory view of an example when the matrix is multistage. (A) is a figure which shows the 1st example of the utilization method of I / O performance statistics. (B) is a figure which shows the 2nd example of the utilization method of I / O performance statistics.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 101 ... Client 102 ... Application program 105 ... Standby queue 106 ... Optimization part 107 ... DBMS 1061 ... Queue monitoring storage part 1062 ... I / O performance storage part 1063 ... I / O matrix replacement part

Claims (6)

A request response that is a response to an I / O request when an I / O request is received from an I / O request issuer, accumulated once in a queue, and predetermined processing is completed for the I / O request output from the queue A computer program for causing a computer to optimize the processing order of I / O requests that can be applied to an I / O control system that returns an I / O request to the I / O request issuer,
A queue status check step for checking how many I / O requests are arranged in what order in the queue and which storage area is the I / O destination of each I / O request;
Based on the result of the check, an I / O destination for each of two or more storage areas that are set as I / O destinations in a plurality of I / O requests arranged in the queue among the plurality of storage areas. The number of I / O requests, which is the number of I / O requests, is calculated, and a plurality of I / O requests in the queue are arranged so that the I / O requests that have a storage area with a large number of I / Os are in the earlier order. A computer program that causes a computer to execute an order rearranging step for rearranging the order of I / O requests. The predetermined process is an I / O command issued to a storage device corresponding to a storage area that is an I / O destination in an I / O request output from the queue, and a response to the I / O command. Receiving a command response from the corresponding storage device;
For each I / O command issued by the I / O control system, the length of time from when the I / O control system issues an I / O command to the storage device until the command response is received from the storage device I / O performance accumulating step for acquiring I / O performance, and accumulating the acquired I / O performance in the storage means,
The computer program according to claim 1, further causing the computer to execute. Based on the plurality of I / O performances stored in the storage means, I / O performance statistics are obtained for each storage device, and the difference in I / O performance between the same types of storage devices exceeds a predetermined value. A storage device optimization step for determining whether or not the storage device is large, and in the case of being large, performing an optimization process on a storage device having a lower I / O performance among the storage devices of the same type;
The computer program according to claim 2, further causing the computer to execute. In the order rearranging step, the plurality of I / O performances stored in the storage unit and the two or more storages that are set as I / O destinations by the plurality of I / O requests arranged in the queue. And determining the I / O performance of each of the two or more storage areas based on the I / O performance of each of the two or more storage areas in addition to the obtained number of I / Os. The priority order of each of the two or more storage areas is determined, and the order of the plurality of I / O requests is arranged so that an I / O request having a storage area having a higher priority order as an I / O destination is earlier. Change
The computer program according to claim 2. In the order rearranging step, from the result of the check, among the plurality of I / O requests arranged in the queue, two or more probabilities that the storage areas as I / O destinations overlap each other are not more than a predetermined value. I / O requests are specified, and two or more specified I / O requests are defined as one I / O with two or more storage areas specified by the two or more I / O requests as I / O destinations. Composite to request,
The computer program according to claim 1. A request response that is a response to an I / O request when an I / O request is received from an I / O request issuer, accumulated once in a queue, and predetermined processing is completed for the I / O request output from the queue Is an I / O control system that returns to the I / O request issuer,
A queue status check unit for checking in which order and how many I / O requests are arranged in the queue and which storage area is the I / O destination of each I / O request;
Based on the result of the check, an I / O destination for each of two or more storage areas that are set as I / O destinations in a plurality of I / O requests arranged in the queue among the plurality of storage areas. The number of I / O requests, which is the number of I / O requests, is calculated, and a plurality of I / O requests in the queue are arranged so that the I / O requests that have a storage area with a large number of I / Os are in the earlier order. An I / O control system comprising an order rearranging unit for rearranging the order of I / O requests.

Images