JP5415338B2 - Storage system, load balancing management method and program thereof - Google Patents

Description

  The present invention relates to a computer system including a plurality of file servers that provide file services to computers, and a storage device system (hereinafter referred to as “storage system”) that provides storage areas to these file servers. The present invention also relates to a management method and a program for realizing load distribution on the system.

  In recent years, with the rapid increase in the amount of data handled in a data center, the access load on the storage system has increased. In particular, online transaction systems and HPC (High Performance Computing) systems input and output large volumes of data. For this reason, the access load on the file server is increasing. As a result, the file server is required to have a high-performance file service that can cope with a high access load.

  Many data centers provide high-performance file services by operating a plurality of file servers in parallel in response to this high performance requirement. Many data centers also provide a shared file system name space that unifies multiple file servers. Thereby, the client can receive a file service as if a plurality of file servers provide one shared file system.

  However, in the conventional system, the load is not evenly allocated to the file servers, and the load may be biased to a small number of file servers. In this case, the performance of the entire shared file system is limited to the performance of some file servers where the load is concentrated. As a result, there is a problem that the read / write speed of the file decreases and the response time increases, and the performance of the file service by the shared file system decreases.

  In order to solve the above-described problem, Japanese Patent Application Laid-Open No. 2004-151867 discloses a technique for distributing client file access to a plurality of file servers. In the case of the method shown in Patent Document 1, a session management unit and a server selection unit installed between a client and a server decompose the file access of the client one by one, and further to the server selected by the round robin method Send file access. This distributes the load on the file server.

  Further, as another means for solving the above problem, Patent Document 2 discloses a technique for moving a part of a file system managed by a file server to another file server. When the file management program for monitoring the state of access from the client to the file server knows that the file access of the client is concentrated on a certain file server, the method of Patent Document 2 manages the file system managed by the file server. Move a part of to another file server. This distributes the load on the file server.

JP 2002-351760 JP 2004-139200 A

  Japanese Patent Application Laid-Open No. 2004-151899, which is a method for distributing the load on a storage system, realizes distribution of a load on a file server by distributing file access between a client and a file server. However, if the distributed file server does not manage the target file or file system, it is necessary to redirect file access to the other file server that is actually managed, and file service is due to that overhead. There was a problem that performance deteriorated.

  The technique of Patent Document 2 realizes load distribution of a file server by moving a part of the file system from a file server where file access is concentrated to another file server. However, there is a problem that it is difficult to move the file system when the load on the file server is high due to concentration of access, for example, when the CPU usage rate is close to 100%. In addition, this technology should move the file access for load distribution, but conversely places a load on the file server. For this reason, there is a problem that the performance degradation of the file service occurs.

  In order to solve the above-described problems, an embodiment of the present invention has the following configuration. Specifically, one or more file servers connected to a plurality of computers to which the first management device is connected, a storage device connected to the file server and having one or more volumes, the file server and the first management A storage system having a second management device connected to the device, wherein the first management device is information on jobs (job information) and job queue information (job queue information) executed sequentially on the computer. , An execution queue, and a queue for waiting for a job until execution.

  The second management device includes means for collecting job information, means for collecting job queue information, means for analyzing the collected job information and job queue information, means for managing load, and a file system. Means for managing. The means for analyzing the job information and the job queue information specifies a file, a file system, and a file server to be accessed by the job based on the job information.

  The means for managing the load calculates the predicted load based on the job information and the job queue information and executes load distribution. The means for managing the file system moves the file system from a server with a high predicted load in the file server to a server with a low predicted load according to the load distribution determined by the means for managing the load.

  In addition, the problem which this application discloses and the solution method are clarified by the column and drawing of embodiment of invention.

  According to the present invention, in a batch processing system application that requires high performance, even when the load changes dynamically and suddenly, the load is distributed between the file servers operating in parallel based on the actual load and the prediction. This makes it possible to provide a storage system that performs high-performance file services.

It is a figure which shows the structural example of the storage system of this invention, a computer connected to it, and a management server. It is a figure which shows the example of a structure of the computer management server and storage management server of this invention. FIG. 2 is a diagram in which a configuration example of a shared file system is extracted from the storage system of FIG. 1. It is a figure which shows the example of the queue before performing the load distribution of a file server, a computer, a file server, a file system, and an estimated load. It is a figure which shows the example of the load distribution by the mount switching of a file system. It is a figure which shows the example of the queue after performing the load distribution of a file server, a computer, a file server, a file system, and prediction load. It is a figure which shows the example of the event which generate | occur | produces in a queue and an execution queue. It is a figure which shows an example of the procedure of the determination of the load distribution timing of a file server of this invention, and load distribution execution. It is a figure which shows an example of the procedure of preparation of the prediction load of this invention. It is a structural diagram showing an example of a load distribution target list and a load list of the present invention. It is a structure figure which shows an example of the file system management table and file server management table of this invention. It is a figure which shows an example of the procedure which calculates threshold value s1. It is a figure which shows an example of the procedure which calculates threshold value s2. It is a figure which shows the example of a computer execution script.

  Embodiments of the invention will be described below with reference to the drawings. The embodiment described below is an example, and the present invention includes a system configuration in which any function described in this specification is combined, or a system configuration in which all or some of the functions described in this specification are combined with a well-known technique. Is also included. The functions executed in the embodiments described later will be described as being realized as a program executed on a computer (computer). However, part or all of the program may be realized through hardware.

[Example 1]
FIG. 1 is a diagram illustrating a configuration of a system including the storage apparatus according to the first embodiment. The computer system 1 includes a computer 11, an IP switch 2, a storage system 9, and a computer management server 7. The storage system 9 includes a file server 3, a fiber channel (FC) switch 4, a storage device 5, and a storage management server 8.

  As shown in FIG. 1, the computer 11 is connected to the storage system 9 by connecting the computer 11 and the file server 3 via the IP switch 2. The computer 11, the file server 3, the storage device 5, and the storage management server 8 are connected to each other via a LAN 6 (Local Area Network) that is a management network.

  As an interface for connecting the file server 3 and the storage apparatus 5, it is common to use an interface of a protocol for sending block data such as fiber channel and iSCSI. Here, the file server 3 and the storage device 5 may be directly connected, but in FIG. 1, they are connected via the FC switch 4.

  The storage device 5 includes a controller 51 and a hard disk mounting unit 58 having a hard disk 59 inside. The controller 51 is connected to a hard disk 59 and a CHA (channel adapter) 54 that is an I / F that controls a data write / read command from a host device such as a file server or a computer. A DKA (disk adapter) 56 that is a disk I / F to be controlled, a cache memory 52, a shared memory 53, a SW 55, an internal LAN 57, and a management terminal 60 are included.

  The cache memory 52 and the shared memory 53 are memory devices shared by the CHA 54 and the DKA 56. The shared memory 53 is mainly used for storing control information and commands. The cache memory 52 is mainly used for storing data.

  The SW 55 connects the cache memory 52, the shared memory 53, the CHA 54, and the DKA 56 to each other. Command or data transmission / reception among the cache memory 52, shared memory 53, CHA 54, and DKA 56 is performed via the SW 55. The SW 55 is generally composed of one or more switch devices that perform data transmission by high-speed switching, for example. However, the SW 55 may be composed of one or more common buses.

  The hard disk mounting unit 58 has one or more groups of hard disks 59 constituting a RAID (Redundant Arrays of Inexpensive Disks). This group of hard disks 59 is called a RAID group (RAIDGr.57). (Modification 1) In the storage apparatus, a logical volume that combines a single or a plurality of RAIDGr57 storage spaces is set. The storage device provides a logical volume as a storage area to the host device. The host device issues a data write / read command for this logical volume.

  The CHA 54 controls data transfer with the cache memory 52 when receiving a data write / read command from the host device. The DKA 56 controls data transfer with the cache memory 52 when writing / reading data to / from the hard disk 59. At this time, the DKA 56 converts the data access request by the logical address designation transmitted from the CHA 54 into a data access request by the physical address designation, and writes / reads data to / from the hard disk 59. By such data exchange between the CHA 54 and the DKA 56 via the cache memory 52, data is written / read to / from the hard disk 59 from the host device. In order to perform such control, the CHA 54 and the DKA 56 have one or more processors (not shown).

  The CHA 54, DKA 56, and management terminal 60 are connected via an internal LAN 57. Further, the storage management server 8 arranged outside the storage apparatus is connected to the internal LAN 57 via the LAN 6. The administrator can set the logical volume, CHA 54, and DKA 56 by operating the management terminal 60 through an input device (not shown).

  Here, the configuration of the controller 51 described above is merely an example, and the configuration is not limited to the above. There is no problem if the controller 51 has a function of writing / reading data to / from the hard disk 59 in response to a data write / read request from the computer 11.

  The computer management server 7 includes a CPU 71, a memory 72, a job scheduler 73, and an IP interface 74. The job scheduler 73 is executed by the CPU 71 using a partial area of the memory 72 and manages jobs executed by the computer 11.

  The storage management server 8 includes a CPU 81, a memory 82, an information collection unit 83, an information analysis unit 84, a load management unit 85, a file system management unit 86, and an IP interface 87. The information collection unit 83, the information analysis unit 84, the load management unit 85, and the file system management unit 86 are executed by the CPU 81 using a partial area of the memory 82. The storage management server 8 executes load balancing of the file server 3 using these functions.

  FIG. 2 shows functions of the computer management server 7 and the storage management server 8. The job scheduler 73 of the computer management server 7 includes a job management unit 201 that manages jobs 211 to be executed by the computer 11, a queue 212 that waits for jobs until execution, and an execution queue that stores jobs being executed by the computer 213, a computer management unit 221 that manages job information (job information) and job queue information (job queue information). The computer management unit 221 performs management such as job execution and termination instructions, job execution status monitoring, and job execution result acquisition to the computer.

  The jobs 211 in the queue 212 are arranged in the job execution order according to the management of the job scheduler 73.

  The storage management server 8 includes an information collection unit 83, an information analysis unit 84, a load management unit 85, and a file system management unit 86. The information collection unit 83 collects information on the job 211 (job information), information on the queue 212 and execution queue 213 (job queue information) from the job scheduler 73, and analyzes the information collected by the information analysis unit 84.

  The load management unit 85 includes a load distribution timing determination unit 231 that determines the timing for executing load distribution, a predicted load creation unit 232 that predicts the maximum load until a certain time later, and a load that executes load distribution based on the predicted load A distribution execution unit 233, a file server management table 234, a load distribution target list 235, and a load list 236 are included.

  The file system management unit 86 includes a file system management table 241 for managing the file system of the file server 3 and a file system migration unit 242.

  The load distribution execution unit 233 instructs the file system management unit 86 to move the file system before the load concentration actually occurs, and the file system movement unit 242 moves the file system between the file servers. Specifically, the file system managed by the high load file server is moved to the low load file server. Thereby, load distribution based on prediction is executed. A specific method for moving the file system will be described later.

  FIG. 3 shows functions of the computer 11 and the file server 3. The computer 11 includes a job execution unit 303 that executes a job received from the job scheduler 73, and executes a calculation program 304 specified by the job. The calculation program 304 performs file access to the file server 3. The file server 3 manages the file system 322 in the volume 321 provided as a storage area by the storage apparatus 5. Further, the file server 3 converts the received file access into block access, and accesses the file system.

  FIG. 4 shows an example of creating a predicted load. The following processing is executed by the load management unit 85. In the example, jobs (JOB) # 1, # 2, and # 3 in the execution queue 213 are being executed, and JOB # 4, # 5, # 6, # 7, and # 8 in the queue 212 are waiting. . As shown in the figure, JOB # 1 to JOB # 8 access file system (FS) # 1 to FS # 8, respectively. NAS # 1 has FS # 1, # 4, # 5, # 6, # 8, NAS # 2 has FS # 2, # 3, # 7, # 9, NAS # 3 has FS # 10, # 11 I manage. The predicted load 403 includes JOBs # 1, # 2, and # 3 waiting in the execution queue 213, and JOBs # 4, # 5, and # 5 having a smaller order in the queue 212 than the threshold s2 401 calculated by a predetermined method. # 6 is the total load applied to each file system FS # 1, # 2, # 3, # 4, # 5, # 6 for each NAS.

  As shown in the figure, the load 405 given by the JOB to the FS # 1, # 2, # 3, # 4, # 5, # 6 is determined according to the order of the jobs 211 in the execution queue 213 and the queue 212. , 3,4,5,6. When the predicted load 403 calculated for each NAS is larger than a predetermined load threshold 404, the load management unit 85 determines that the NAS has a high load. In the figure, only the predicted load 403 of NAS # 1 is higher than the load threshold 404. At this time, the load management unit 85 determines that NAS # 1 has a high load.

  FIG. 5 shows an example of a method used when the file system moving unit 86 moves a file system between file servers. In the example, NAS # 1 501 provides a file service to the computer 11. Further, the computer 11 NFS mounts the NFS share 511 on the directory 521. In this case, the computer 11 can access the file as if the file system 322 in the volume 321 is in the computer 11 by accessing the directory 521.

The file system is moved in the following four procedures.
(1) The file system moving unit 242 issues an instruction to the computer 11. In accordance with the instruction, the computer 11 unmounts the NFS share 511 by NFS.
(2) The file system moving unit 242 issues an instruction to the NAS # 1 501. According to the instruction, the NAS # 1 501 stops the NFS share 511, and then unmounts the file system 322.
(3) The file system moving unit 242 issues an instruction to the NAS # 3 502. In accordance with the instruction, the NAS # 3 502 mounts the file system 322 as a file system, and then starts the NFS sharing 512.
(4) The file system moving unit 242 issues an instruction to the computer 11. In accordance with the instruction, the computer 11 NFS mounts the NFS share 512 on the directory 521.

  As a result of the movement of the file system, NAS # 3 manages the file system 322 instead of NAS # 1, and NAS # 3 provides a file service when the computer 11 accesses the directory 521.

  In the case of the above example, in the procedure (1) and the procedure (4), the file system moving unit 242 directly instructs the computer 11 to mount the NFS unmount. However, the file system moving unit 242 employs a technique for controlling the NFS unmount and mount of the computer 11 by giving an instruction to a system that shares information between computers on a network such as NIS (Network Information Service). Also good.

  FIG. 6 shows an example of a load distribution execution operation based on the predicted load created in FIG. 4 and changes in the predicted load. In the load distribution execution, the operation shown in FIG. 5 is performed. Before that, first, the source and destination file servers of the file system and the target file system are selected. In the case of FIG. 4, NAS # 1 is determined to have a high load, and the load 6 exceeds the load threshold 404.

  In this case, the file server with the highest predicted load, that is, NAS # 1 in FIG. 6, is selected as the migration source of the file system in load distribution.

  Also, a load exceeding the load threshold 404, that is, the file system FS # 6 with the load 6 in FIG. 6 is selected as the target file system for load distribution.

  Further, the file server with the smallest predicted load value, that is, NAS # 3 in FIG. 6 is selected as the destination of the file system.

  Next, the file system is actually moved based on the file system move operation shown in FIG. 5 based on the determined file system move source, move destination file server, and target file system information. That is, the load distribution in FIG. 6 is realized by moving the file system FS # 6 of NAS # 1 to NAS # 3 by the method shown in FIG. This operation is executed in advance before JOB # 6 corresponding to the load 6 starts execution. With this movement of the file system, the load 6 moves from NAS # 1 to NAS # 3. For this reason, the predicted load of NAS # 1 is suppressed to be smaller than the load threshold 404. As a result, even when the load dynamically changes due to job execution, it is possible to perform load distribution between file servers in advance before job execution.

  Always performing load balancing should be avoided to generate load on the file server. By monitoring the events occurring in the queue 212 and the execution queue 213 and executing load distribution only when the event occurs, the load on the file server can be reduced.

  FIG. 7 shows events that occur in the queue 212 and the execution queue 213 used to determine the execution timing of load distribution. The job statuses in the queue 212 and the execution queue 213 are collectively referred to as a job status.

  First, in the job state (t = t1), the execution queue 213 has only one job # 2, and the queue 212 has jobs # 3, # 4, # 5, # 6, and # 7.

  When the job status (t = t2) is reached, JOB # 3 214 is moved from the queue 212 to the execution queue 213. This means that execution of JOB # 3 215 has started on the computer 11. A job execution start event is detected from changes in the job state (t = 2) and (t = 1).

  When the job status (t = t4) is reached, the job # 6 216 in the queue 212 that existed in the job status (t = t3) has disappeared. This means that JOB # 6 has been canceled by the user, and the scheduler has deleted JOB # 6 from the queue 212. A job cancel event is detected from changes in the job state (t = 2) and (t = 1).

  Next, when the job state (t = t6) is reached, JOB # 2 217 existing in the queue 213 disappears in the job state (t = t5). This means that JOB # 2 has finished executing. A job execution end event is detected from changes in the job status (t = 6) and (t = 5).

  However, the load distribution execution timing is not limited to the above-described method. For example, (1) The load is distributed every time the event occurs for a preset number of times, (2) The load is distributed every time the total load change due to the occurrence of the event becomes greater than a preset value, (3) Preset For example, there is a method of distributing the load every time.

  The above-mentioned load means a load given to the file server by the job accessing the file system. For example, a file read / write transfer speed, IOPS (input output per second), or a CPU utilization rate in a NAS file service can be defined as a load. The load management unit 85 uses these values as the magnitude of the load. A value that gives the magnitude of the load is input to the load magnitude 1027 in the load list 1003 managed in the load manager 85 and used for the load distribution process. In addition to this definition, a value calculated by some method such as adding a weight to the above values can be defined as a load.

  In each file server, a load threshold 404 that is an upper limit of the load is set in advance. The load threshold 404 is set in advance as the load threshold 1122 of the file server management table 1102 and is used for the load distribution process.

  FIG. 8 shows a load distribution procedure in the file server of the present invention, and the load management unit 85 executes the processing. First, in step 801, the load management unit 85 collects job statuses from the job scheduler 73. In the next step 802, the load management unit 85 analyzes the job status and detects an event. The event detection method is shown in FIG. 7 as described above.

  In step 803, the load management unit 85 determines whether the generated event is a load distribution target event. If the generated event is a job execution start, job cancellation, or job execution end event, the load management unit 85 proceeds to step 804. Otherwise, the load management unit 85 returns to Step 801.

  In step 804, the load management unit 85 executes load distribution. The detailed contents of load distribution will be described below.

  First, the load management unit 85 creates a predicted load in Step 811. This procedure will be described in detail with reference to FIG. Next, the load management unit 85 creates a load list 1001 in which loads exceeding the load threshold 404 are collected in step 812. The load list 1001 exists in the load management unit 85 and is managed by the load management unit 85.

  In subsequent step 813, the load management unit 85 deletes the load satisfying the queue order 1014 ≦ threshold s 1 402 from the load list 1001. A method for determining the threshold value s1 will be described in detail with reference to FIG. In the next step 814, the load management unit 85 determines whether the size of the load list 1001 is 1 or more. When the size is 0, the load management unit 85 ends the load distribution process. If the size is 1 or more, the load management unit 85 proceeds to step 815.

  In step 815, the load management unit 85 selects the load distribution target 1002 having the smallest mount source NAS 1015 number described in the load distribution target list 1001. When there are a plurality of smallest load distribution targets 1002, the load distribution target 1002 having the smallest queue order 1014 is selected. In step 816, the NAS # b having the smallest predicted load 403 is selected as the file system migration destination for load distribution.

  In step 817, the load management unit 85 sets the file system 1013 described in the load distribution target 1002 selected in step 815 as the load distribution target, NAS (1015) as the mount source, and NAS # b as the mount switching destination. A mount switching instruction is issued to the system moving unit 242. As described above, the mount switching is performed by the file system moving unit 242 of the file system managing unit 86 according to the method shown in FIG. When step 817 ends, the load management unit 85 returns to step 811.

  FIG. 9 shows a procedure for creating the predicted load 403 in step 811. This process is executed by the predicted load creation unit 232 of the load management unit 85. First, the predicted load creation unit 232 initializes the load list 1003 (236 in FIG. 2) in step 901.

  Next, the predicted load creation unit 232 acquires a job script 1401 for each job from the job scheduler 73 in step 902. In step 903, the predicted load creation unit 232 analyzes each job script 1401 of each acquired job. As a result of the analysis, the predicted load creation unit 232 determines a job access destination file system. Also, the predicted load creation unit 232 uses the file system management table 1101 to determine the mount source NAS from the file system obtained by the determination, and adds it as load data 1004 to the load list 1003.

  In the next step 904, the predicted load creation unit 232 acquires the order in the queue 212 and the execution queue 213 corresponding to the job ID 1022 of each load data 1004 in the load list 1003 from the job scheduler 73, and executes the load list 1003. Input is made to the order 1025 in the queue and the order 1024 in the queue.

  In the next step 905, the predicted load creating unit 232 inputs the load size corresponding to the job ID 1022 of each load data 1004 in the load list 1003 to the load size 1027 of the load list 1003.

  In the next step 906, the predicted load creating unit 232 deletes the load data 1004 satisfying the order 1024 ≧ threshold s 2 401 in the queue from the load list 1003. The method for determining the threshold value s2 will be described in detail with reference to FIG.

  In the next step 907, the predicted load creating unit 232 initializes the predicted load 403 to 0 for each NAS.

  In the last step 908, the predicted load creating unit 232 adds a load 405 corresponding to the load size 1027 to the predicted load 403 corresponding to the mount source NAS 1026 described in each load data 1004. Then, a series of processing ends.

  In the creation of the predicted load described here, the load list 1003 is recreated from initialization every time it is created. By recreating it, you can definitely make a load prediction, but the same process is duplicated in the process, so the efficiency of creating it is not good.

  In particular, when a large number of file systems 322 are managed by the file server 3, the amount of processing necessary for creating a predicted load increases due to re-creation. When the frequency of changing the load is high, the prediction load is created with a high frequency, and thus more processing is required.

  On the other hand, instead of recreating the predicted load, there is also a method of efficiently creating the predicted load by updating the change in the predicted load. For example, when the jobs in the queue 212 are in an order smaller than the threshold s 2, the load corresponding to the jobs is updated by adding to the predicted load 403. When a job execution end event or a job cancel event occurs, the predicted load creating unit 232 updates the job by deleting the load corresponding to the job that has ended from the predicted load 403. However, it is necessary to reliably update the change in the predicted load.

  Therefore, if there is a time when the management server 8 cannot update the load prediction change due to a failure or the like, it can be dealt with by initializing and recreating the load prediction. In addition, there is also a method for periodically initializing and regenerating the load prediction in order to reliably create the load prediction.

  FIG. 12 shows a procedure for calculating the threshold value s1 402. This process is also executed by the load management unit 85. First, the time at which the job execution event occurs in step 1202 is totaled as a sample.

  Next, the probability P1 of how many jobs are executed within the time t1 is calculated based on the sample of job execution event occurrence times counted in step 1203. For example, when a sample occurs from time ts to time te, the probability P1 (1) that one or less job is executed is calculated as follows. First, the number of executions of one or less jobs between time t and time t + t1 is counted from the sample every time t at 1 minute intervals from time ts to time te. Next, the number of times is divided by the number of times t at 1-minute intervals from time ts to time te, and the probability P1 (1) is calculated. In this way, the probability P1 is calculated such that the probability P1 (2) that two or less jobs are executed and the probability P1 (3) that three or less jobs are executed.

  The calculation method may be another method as long as the probability P of how many jobs are executed within the time t1 can be calculated based on the aggregated job execution event occurrence time samples. For example, the distribution of the probability P1 that the job is executed is estimated from a sample of the time at which the job execution event occurs statistically, and the probability that the job occurs within the time t1 may be calculated from the estimated distribution.

  An example of the probability P1 is shown in a graph 1210 in FIG. Next, in step 1204, the maximum value in the order k in the queue where probability P1> threshold value Pth1 is calculated. In the figure, k = 1. In step 1205, k calculated in step 1204 is set to the threshold s1. By setting the threshold value s1, the load of the job in the queue 212 in order k ≦ s1 is excluded from the candidates for load distribution, and the job execution is not started while the file system is moving.

  Therefore, the time t1 sets the time required for moving the file system. For example, there is a method of setting a maximum value of the file system movement time determined by the specification of the file server or a value obtained by adding a margin of several minutes to the maximum value of the movement time when the file system has been moved so far.

  The threshold value P1th sets a probability that job execution may be started during file system migration. For example, a sufficiently small probability such as threshold value P1th = 0.0001 is set. This threshold s1 is called a load distribution selection threshold.

  FIG. 13 shows a procedure for calculating the threshold s2 401. This process is also executed by the load prediction unit 85. First, in step 1302, the time when the job execution event occurs is counted as a sample.

  Next, the probability P2 of how many jobs are executed within the time t2 is calculated based on the sample of job execution event occurrence times counted in step 1303. For example, when a sample is generated from time ts to time te, the probability P2 (1) that one or less job is executed is from time t to time t every 1 minute interval from time ts to time te. Count the number of times one or less jobs were executed during t + t2 from the sample.

  Then, the number of times can be calculated by dividing the number of times by the number of times t at 1 minute intervals from time ts to time te. In this way, the probability P2 is calculated as the probability P2 (2) that two or less jobs are executed and the probability P2 (3) that three or less jobs are executed.

  As the calculation method, other methods may be used as long as the probability P2 of how many jobs are executed within the time t2 can be calculated based on the sampled occurrence times of job execution events. For example, the distribution of the probability P2 that the job is executed is estimated from a sample of the time when the job execution event occurs statistically, and the probability that the job is generated within the time t2 may be calculated from the estimated distribution. An example of the probability P2 is shown in a graph 1310 in FIG.

  Next, in step 1304, the smallest one in the order k in the queue where probability P2 <threshold value Pth2 is calculated. In the figure, k = 5. In step 1305, k calculated in step 1304 is set to the threshold s2. By setting the threshold value s2, it can be determined that the job in the queue 212 in the order k ≦ s2 has a long time until job execution.

  Then, by excluding the job load from the predicted load, the accuracy of the job execution predicted load is increased. Therefore, the time t2 sets a time range in which the predicted load is used for calculation. Since the predicted load is calculated by collecting the loads of jobs that may be executed at the same time, for example, the maximum execution time among jobs executed so far is set as the time t2.

  If it is considered that the maximum execution time of the job is abnormally long only at that time and cannot be relied upon, for example, the second longest time may be set as the execution time of the job. The threshold value s2 sets a probability that a job that is predicted to exceed the time range used for calculating the predicted load may be executed contrary to the prediction. For example, a sufficiently small probability such as threshold value P2 = 0.0001 is set. This threshold value s2 is called a predicted load exclusion threshold value, and time t2 is called a predicted load target time.

  FIG. 10 shows an example of the structure of the load distribution target list 1001 (235 in FIG. 2) and the load list 1003 (236 in FIG. 2). The load distribution target list 1001 includes a load management number 1011, a job ID 1012, a job access destination file system 1013, an order 1014 in the queue corresponding to the job ID, a NAS 1015 that mounts the file system, and a load size 1016. Is done. The load distribution target 1002 is input to the list row. The load list 1003 includes a load management number 1021, a job ID 1022, a job access destination file system 1023, an order 1024 in the queue corresponding to the job ID, an order 1025 in the execution queue corresponding to the job ID, and a file system mount source NAS 1026 and the load size 1027 corresponding to the job ID.

  FIG. 11 shows an example of the structure of the file system management table 1101 (241 in FIG. 2) and the file server management table 1102 (234 in FIG. 2). The file system management table 1101 includes a file system name 1111, a directory name 1112 corresponding to the directory 521 in which the file system is NFS mounted on the computer side, and a mount source NAS 1113. The file server management table 1102 includes a NAS name 1121 and a load threshold value 1122.

  FIG. 14 shows an example of the contents of the job script 1401. The job script 1401 includes a CPU count 1411 used by the job, a maximum job execution time 1412, a maximum memory count 1413 used by the job, an input file name 1414 read by the job as input data, and an output file name to which the job outputs data. 1415, the job has an execution file name 1416 of the program executed by the computer 11.

1 Computer system 2 IP switch 3 File server 4 FC switch 5 Storage device 6 LAN
7 Computer management server 8 Storage management server 51 Controller 57 RAIDGr.
58 hard disk mounting unit 59 hard disk 73 job scheduler 83 information collecting unit 84 information analyzing unit 85 load managing unit 86 file system managing unit

Claims (11)

Two or more file servers connected to a plurality of computers to which a first management device is connected, a storage device connected to the file server and having one or more volumes, the file server, and the first management device; A storage system having a second management device connected to the storage system,
The first management device includes an area for storing job information (job information) and job queue information (job queue information) sequentially executed on a computer, an execution queue, and a queue for waiting for a job until execution. Have
The second management device includes means for collecting the job information, means for collecting the job queue information, means for analyzing the collected job information and the job queue information, means for managing a load, Means for managing the file system,
The means for analyzing the job information and the job queue information specifies a file, a file system, and a file server to be accessed by the job based on the job information,
The means for managing the load calculates a predicted load for each file server based on the job information and the job queue information, and determines that load distribution is performed when the predicted load exceeds a predetermined load threshold. ,
The means for managing the file system, when the means for managing the load determines to execute load distribution , changes the file system corresponding to the load from a file server having a high predicted load in the file server to a low predicted load. A storage system that moves to a file server. The storage system according to claim 1, wherein
The means for managing the load selects a file server that exceeds a preset load threshold from among the predicted loads calculated for each of the file servers, and the load threshold is exceeded. storage system, wherein a partial load is selected as load balancing target, selects the smallest file server estimated load as a file server of the destination load. The storage system according to claim 2, wherein
The means for managing the load specifies a file system and a file server to be accessed by a job arranged in the job queue based on the job information and the job queue information, and sets each file arranged in the job queue. A storage system characterized in that a predicted load is calculated by summing up the load given by a job for each of the file servers. The storage system according to claim 3, wherein
The storage system, wherein the load managing unit deletes the load from the load distribution target when the order of jobs corresponding to the load in the job queue is equal to or less than a load distribution selection threshold. The storage system according to claim 4, wherein
The storage system, wherein the means for managing the load deletes a load given by a job whose order in the job queue is equal to or higher than a predicted load exclusion threshold. The storage system according to claim 5, wherein
Said means for managing the load on the basis of the frequency of events occurs, calculates a maximum number of times a predetermined method to perform job in a range of a predetermined probability within a predetermined estimated load target time The storage system is characterized in that the number of times is set as a predicted load exclusion threshold. The storage system according to claim 6, wherein
The storage system characterized in that the means for managing the load detects an event occurring in the job queue by taking a difference between current and past contents of the job queue information. The storage system according to claim 7, wherein
The storage system characterized in that the means for managing the load determines load distribution timing by a predetermined method based on the occurrence of the event. The storage system according to claim 7, wherein
The means for managing the load calculates a maximum number of times that the job is executed within a predetermined probability range within a predetermined file system movement time based on the frequency of occurrence of the event by a predetermined method. The number of times is set as a load balancing selection threshold. Two or more file servers connected to a plurality of computers to which a first management device is connected, a storage device connected to the file server and having one or more volumes, the file server, and the first management device; A load distribution management method in a storage system having a second management device connected to
The first management device includes an area for storing job information (job information) and job queue information (job queue information) sequentially executed on the computer, an execution queue, and a queue for waiting for a job until execution. If you have
The second management device includes a process for collecting the job information, a process for collecting the job queue information, a process for analyzing the collected job information and the job queue information, a process for managing a load, Process to manage the file system,
The process of analyzing the job information and the job queue information specifies a file, a file system, and a file server that the job accesses based on the job information,
The process for managing the load calculates a predicted load for each file server based on the job information and the job queue information, and determines that load distribution is executed when the predicted load exceeds a predetermined load threshold. ,
The process of managing the file system, when the means for managing the load decides to perform load balancing, the file system corresponding to the load, small estimated load from a high file server predictable load of the file server A load balancing management method for a storage system, characterized by moving to a file server. Two or more file servers connected to a plurality of computers to which a first management device is connected, a storage device connected to the file server and having one or more volumes, the file server, and the first management device; A program for causing a computer to execute load balancing management processing in a storage system having a second management device connected to
The first management device includes an area for storing job information (job information) and job queue information (job queue information) sequentially executed on the computer, an execution queue, and a queue for waiting for a job until execution. If you have
The computer mounted on the second management device has a process of collecting the job information, a process of collecting the job queue information, a process of analyzing the collected job information and the job queue information, and a load. Execute the process to manage and the process to manage the file system,
The process of analyzing the job information and the job queue information specifies a file, a file system, and a file server that the job accesses based on the job information,
The process for managing the load calculates a predicted load for each file server based on the job information and the job queue information, and determines that load distribution is executed when the predicted load exceeds a predetermined load threshold. ,
The process of managing the file system, when the means for managing the load decides to perform load balancing, the file system corresponding to the load, small estimated load from a high file server predictable load of the file server A program that enables a computer to implement load balancing management of a storage system, characterized by moving to a file server.

Images