JP5696891B2 - Job scheduling method by IO configuration - Google Patents

Description

  The present invention relates to high-speed processing of large-scale data.

  Conventionally, the amount of IO data has been increased due to the increase in CPU speed and memory capacity. In particular, in the field of HPC (High Performance Computing), the importance of IO has increased as file sharing in a large-scale environment has become common, and it is also necessary to consider IO from the viewpoint of job management. Desired. However, the conventional job scheduler mainly targets the number of CPUs, the amount of memory, and the like as management targets, and does not take into account the IO configuration, and therefore cannot effectively extract the IO performance of the system. Patent Document 1 is an example of a job scheduling method.

JP 2008-242614 A

  Computer systems are generally improved in performance by increasing the number of nodes rather than the size of each node. As a configuration of such a computer system, there can be considered a configuration in which there are a large number of switches in which nodes and disk devices are connected, and each node accesses each disk device via these switches. Note that the switches are connected to each other or by a higher-order switch. In such a computer system configuration, when focusing on IO performance, job input files are often placed in a specific area such as the user's home area, and are connected to the same switch as the file storage disk device. Should be assigned to the job.

  However, in the conventional computer system, when an empty node is assigned to a job, there is no means for reflecting the IO configuration, and it is impossible to assign a node effective for IO performance. For this reason, a problem that the execution time of a job having a large amount of IO is particularly long has been pointed out. Even when the same job is executed, the execution time may vary depending on the node on which the job operates.

  In view of such a situation, the present invention has a large amount of IO in a field where large-scale data such as HPC (High Performance Computing) must be processed at high speed, and particularly when large-scale file sharing is performed. It is intended to provide available job scheduling.

  One of the computer systems according to the present invention is a data file used in a job in consideration of the IO configuration of the system when scheduling a job executed in the computer system in a computer system in which a large number of nodes are connected by a network line. A node in which the IO path is close to a disk device (hereinafter referred to as a “file storage disk device”) in which (job input file) is stored is assigned to the job to ensure IO performance. In addition, when only a node with a long distance can be secured, the job input file is copied to a disk device close to the secured node, thereby having a function of securing IO performance.

  According to the job scheduling of the present invention, firstly, the effect that the execution time of the job can be shortened can be obtained. This is because the IO time is shortened by performing file access from a node close to the disk device.

  Second, it is possible to achieve an effect that variation in job execution time is reduced. Conventionally, the nodes assigned at the time of job execution are irrelevant to the IO configuration, and even if the same job is executed, the execution time varies depending on the distance between the assigned node and the file storage disk device. This can be improved by assigning nodes according to the IO configuration. In particular, in the present invention, when the distance is long, a file is copied to a closer disk, and therefore, in the case of assigning the node with the longest distance, a great effect is seen compared to the conventional method.

It is a block diagram which shows schematic structure of the computer system which concerns on one Embodiment of this invention. It is a block diagram which shows schematic structure of the job server 1 in a present Example. It is a figure for demonstrating an example of the apparatus structure of the computer system by a present Example, and its effect. 6 is a flowchart showing a flow of processing for securing a node in the present embodiment. In this embodiment, it is a flowchart showing a flow of processing to secure an empty node closest to the switch. 5 is a flowchart illustrating a flow of processing for releasing a node in the present embodiment.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In addition, the same code | symbol is attached | subjected to the same element and the overlapping description is abbreviate | omitted.

  FIG. 1 is a block diagram showing a schematic configuration of a computer system according to an embodiment of the present invention. As shown in the figure, in this embodiment, the job server 1, the file server 4, the first identical switch connection device 31, and the second identical switch connection device 32 are connected to each other via the network line 2. ing. The first same switch connection device 31 and the second same switch connection device 32 are device groups connected to the switches 311 and 321, respectively, and a node, a disk device, and a file system created on the disk device, respectively. including. The file system is shared by all nodes, and the file server 4 manages the file system. A file system can be composed of a plurality of disk devices connected to the same switch, but is not composed of disk devices connected to different switches.

  This embodiment is based on the premise that a large number of the same switch connection devices are connected by the network line 2. Each node can access not only the disk device in the same switch connection device including itself but also the disk device in the other same switch connection device through the network line 2, and does not pass through the file server 4. Access data. The job server 1 and each node can access the same file with the same path name. The operating system is premised on UNIX (registered trademark), Linux, and the like.

  FIG. 2 is a block diagram illustrating a schematic configuration of the job server 1 in the present embodiment. As shown in FIG. 1, a job scheduler 11 that controls a job submitted by a user operates on the job server 1. The job scheduler 11 includes a system configuration information file 101, a system configuration information registration unit 102, a node reservation / release unit 103, a file path information acquisition unit 104, a file system identification unit 105, a switch identification unit 106, a file System / switch correspondence table 107, network configuration information 108, free node selection means 109, switch subordinate node information 110, node information change means 111, file operation means 112, and job scheduler standard function 113 Is done.

  The system configuration information file 101 contains information related to the system configuration such as a threshold of the distance between the file storage disk device and the node, a list of nodes and subordinate file systems connected to the switch for each switch, and a network configuration. It is a described file. It is placed on the job server 1 and created and changed by the system administrator.

  The system configuration information registration unit 102 reads the system configuration information file 101 when the job scheduler 11 is started up, captures a threshold value of the distance between the file storage disk device and the node, a file system / switch correspondence table 107, node information under the switch 110. The network configuration information 108 is created.

  The node securing / releasing unit 103 performs overall control of processing related to allocation and release of nodes to jobs. The file path information acquisition unit 104 acquires the path name of the input file of the job specified by the user. The file system specifying unit 105 specifies the file system in which the file exists from the path name of the file.

  The switch specifying means 106 refers to the file system / switch correspondence table 107 and specifies the switch to which the file storage disk device is connected. Since the file system is not composed of disk devices connected to a plurality of switches, if the correspondence between the file system and the switch is known, the switch to which the file storage disk device is connected can be specified.

  The file system / switch correspondence table 107 shows the correspondence between the disk device and the switch, and is used to specify which switch the file storage disk device is connected to. The network configuration information 108 is information on the connection configuration between the switches, and the distance between the switches can be known by referring to this information.

  The empty node selecting unit 109 is called from the node securing / releasing unit 103 and returns information on empty nodes at a closer distance (an empty node, a switch to which the empty node is connected, a distance) to the file storage disk device. The node securing / releasing means 103 determines whether to copy the file based on this distance.

  The switch subordinate node information 110 is information including a list of nodes connected to each switch and the state of the node (free, in use, in failure, etc.). The node information changing unit 111 is used to update the state of the node included in the switch subordinate node information 110 when a node is secured / released.

  The file operation unit 112 is called from the node securing / release unit 103 when the distance between the secured node and the file storage disk device is equal to or greater than the threshold, and the job input file is transferred to the switch under the switch to which the secured node is connected. Copy to a disk unit. As a result, a file IO having a short distance from the node on which the job operates can be obtained. When the node is released, the copied file is deleted. The job scheduler standard function 113 indicates general functions that the scheduler normally has such as queuing of jobs submitted by the user, sequence control based on the resources requested by the job, job start / end, status monitoring, etc. It is.

  Next, an operation when assigning a node to a job in this embodiment will be described.

  FIG. 4 is a flowchart showing the flow of processing for securing a node in this embodiment. The node reservation / release unit 103 receives the node reservation request from the job scheduler standard function 113, and the file path information acquisition unit 104 acquires the path name of the input file of the job specified by the user (S41). Given the path name of a file, the general function of the operating system can be used to obtain the file system in which the file exists. The node securing / releasing means 103 uses the file system specifying means 105 to specify the file system in which the file exists from the file path name (S42). Subsequently, the switch identification unit 106 is used to identify the switch with the file system based on the file system / switch correspondence table 107 (S43), and the empty node selection unit 109 determines the closest distance from the switch. A node is secured (S44). Next, it is checked whether the distance between the node secured in step S44 and the file storage disk device is equal to or greater than a preset threshold value (S45). The job input file is copied to the disk device under the same switch (S46). The job accesses the copied file. Next, for the reserved node, the node information changing unit 111 is used to update the node state in the switch subordinate node information 110 from “free” to “in use” (S47).

  FIG. 5 is a flowchart showing the flow of processing for securing an empty node at the closest distance from the switch in this embodiment. As shown in the figure, first, the free node selecting unit 109 initially sets the distance between the switch and the file storage disk device to be searched for a free node first (S51). The target switch is the switch specified in step S43 in FIG. 4, and the distance is 1 which is the distance to the node connected to the switch. Next, whether or not there is an empty node in the target switch is checked by the switch subordinate node information 110 (S52). If there is an empty node, an empty node is secured (S54), and if there is no empty node, the network configuration information 108 is used. The next closest switch is set as a target, and the distances are added (S53), and the processes in and after step S53 are repeated until an empty node can be secured. If there is no free node on the computer system, the job scheduler standard function 113 does not make a node reservation request.

  FIG. 6 is a flowchart showing the flow of processing for releasing a node in this embodiment. As shown in the figure, in response to a request from the job scheduler standard function 113, the node securing / releasing unit 103 uses the node information changing unit 111 for the node to be released to determine the node in the node information 110 under the switch. The state is updated from “in use” to “free” (S61). Next, when the file is copied because the distance is equal to or greater than the threshold when the node is secured, the input file of the copied job is deleted by the file operation unit 112 (step S62).

  FIG. 3 is a diagram for explaining an example of the apparatus configuration of the computer system according to the present embodiment and its effect. In the process of securing a free node by the job scheduler, the switch to which the file storage disk device is connected is identified from the path name of the input file of the job so that a node closer to the file storage disk device can be secured. The distance between the disk device and the node is determined by the number of switches that pass through. For example, in the configuration shown in FIG. 3, it is assumed that a job input file is stored in the disk device # 1. In this case, the nodes closest to the disk device # 1 are the node # 1-1 and the node # 1-2 connected to the same switch # 1, and the distance is considered to be 1 because one switch is passed through. Similarly, since the node # 2-1 and the node # 2-2 connected to the switch # 2 pass through three switches, the distance from the disk device # 1 is 3. A node with this small distance is determined as a close node.

  The job scheduler identifies the switch to which the file storage disk device is connected from the correspondence table between the file system and the switch, and allocates a free node connected to the switch. If there is no vacancy in the node connected to the switch, it is checked whether there is a vacancy in the node connected to the next closest switch based on the network configuration information. By repeating this until a free node can be secured, a closer node is assigned.

  However, even when an empty node is assigned in this way, if there is no available node near the file storage disk device, a node far from the file storage disk device may be assigned as a result. In such a case, the IO performance of the job is reduced or the number of switches through which the job passes is increased, so that it is easily affected by external factors other than the job. Therefore, a threshold is set for the distance between the node to be allocated and the file storage disk device, and if this threshold is exceeded, the job input file is copied to the disk device connected to the same switch as the allocation node, and the copied file The above problem is solved by using as an input file.

  The present invention is not limited to the above-described embodiment, and can be implemented in various other forms without departing from the gist of the present invention. For this reason, the said embodiment is only a mere illustration in all points, and is not interpreted limitedly. For example, the above-described processing steps can be executed in any order or in parallel as long as there is no contradiction in the processing contents.

A part or all of the above embodiments can be described as in the following supplementary notes, but is not limited thereto.

  (Supplementary note 1) In a computer system in which a plurality of nodes are connected by a network line, the computer system includes a job server for performing scheduling for assigning a node to a job, and the job server has a data file used in a job. A computer system, wherein a node having a closer IO path distance to a stored disk device is assigned.

  (Supplementary Note 2) If the distance between the allocation path and the IO path of the disk device storing the data file used in the job exceeds a predetermined threshold, the job server determines that the distance between the IO path The computer system according to appendix 1, wherein a node is assigned after copying to a closer disk device.

  (Supplementary Note 3) In a computer system in which a plurality of nodes are connected by a network line, a job server included in the computer system performs job scheduling, and the job server performs scheduling for assigning nodes to jobs. A node having a closer IO path to a disk device storing a data file used in the job.

  (Additional remark 4) When the distance of IO path | route of the disk apparatus in which the data file used by the allocation node and the said job is stored exceeds the predetermined | prescribed threshold value, the said job server will distance the said data file to IO path | route. The method according to appendix 3, wherein a node is allocated after copying to a closer disk device.

DESCRIPTION OF SYMBOLS 1 Job server, 2 Network line, 31, 32 Same switch connection apparatus, 311, 321 switch, 4 File server, 11 Job scheduler, 101 System configuration information file, 102 System configuration information registration means, 103 Node reservation / release means, 104 File path information acquisition means, 105 file system identification means, 106 switch identification means, 107 file system / switch correspondence table, 108 network configuration information, 109 free node selection means, 110 switch subordinate node information, 111 node information change means, 112 file Operation means, 113 Job scheduler standard function

Claims (4)

In a computer system in which multiple nodes are connected by a network line,
The computer system includes a job server for performing scheduling for assigning nodes to jobs,
The job server
A job scheduler for allocating a node having a closer IO path distance to a disk device storing a data file used in a job, the IO of the disk device storing the data file used in the job A job scheduler for allocating nodes after copying the data file to a disk device having a closer IO path distance when the path distance exceeds a predetermined threshold;
The job scheduler , when releasing a node, deletes the copied data file if the data file is copied because the IO path distance exceeds the threshold when the node is allocated. Computer system to do.   The computer system according to claim 1, wherein the job server does not make a node reservation request when there is no empty node on the computer system. In a computer system in which a plurality of nodes are connected by a network line, a job server included in the computer system performs job scheduling,
The job server
When scheduling to assign a node to a job, assign a node with a closer IO path distance to the disk device that stores the data file used in the job,
If the IO path distance between the allocation node and the disk device storing the data file used in the job exceeds a predetermined threshold, after copying the data file to a disk device with a closer IO path distance When allocating a node and releasing a node, if the data file was copied because the IO path distance exceeded the threshold when allocating the node, delete the copied data file. Feature method.   The method according to claim 3, wherein the job server further does not make a node securing request when there is no free node on the computer system.