Switch-aware parallel file system

• US 8,701,113 B2
• Filed: 05/27/2010
• Issued: 04/15/2014
• Est. Priority Date: 05/27/2010
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Hildebrand, Dean, Tewari, Renu
• Primary Examiner(s): Dao, Tuan
• Assistant Examiner(s): Mudrick, Timothy A
• Application Number:
  US12/788,937
• Publication Number:
  US 20110296422A1
• Time in Patent Office: 1,419 Days
• Field of Search: 718/102
• US Class Current: 718/102
• CPC Class Codes: G06F 16/1824 implemented using Network-attached Storage [NAS] architecture distributed or networked storage systems G06F3/067; protocols for distributed storage of data in a network H04L67/1097

Abstract

Embodiments of the invention related to a switch-aware parallel file system. A computing cluster is partitioned into a plurality of computing cluster building blocks comprising a parallel file system. Each computing cluster building block comprises a file system client, a storage module, a building block metadata module, and a building block network switch. The building block metadata module tracks a storage location of data allocated by the storage module within the computing cluster building block. The computing cluster further comprises a file system metadata module that tracks which of the plurality of computing cluster building blocks data is allocated among within the parallel file system. The computing cluster further comprises a file system network switch to provide the parallel file system with access to each of the plurality of computing cluster building blocks and the file system metadata module. At least one additional computing cluster building block is added to the computing cluster, if resource utilization of the computing cluster exceeds a pre-determined threshold.

Claims

1. A computing cluster for reducing inter-switch network traffic, comprising:

two or more building blocks partitioned from said cluster, each building block including;

a network switch that provides said cluster with access to the building block and is connected to other building blocks using a hierarchy of network switches, at least one computing server, at least one storage server, a storage system, and a server job scheduler that is in communication with said computing servers within the same building block;

a parallel file system that provides parallel data access to computing servers within each building block and global access to data across said cluster using said switches, said file system including;

a file system client on each computing server that serves file system application requests and has access to data allocated among each of the building blocks, a storage module on each storage server that stripes data across a storage system within a particular building block the storage module is located and serves data requests for the file system clients in said cluster, and a building block metadata module within each building block that tracks a storage location of data allocated by the storage module to the storage system within each building block, and a global metadata module that maintains file system metadata for said file system and said metadata including which building block data is allocated among within said file system; and

wherein said scheduler schedules a job of said file system within a building block and spawns said job at a scheduled time;

i) on at least one of at least one computing server in the same building block such that job requirements of the job are met, ii) on at least one of said building blocks where data for said job is located if there is pre-existing data for said job in said cluster, or iii) on at least one of said building blocks having most available resources if there is no pre-existing data for said job in said cluster.

2. The computing cluster of claim 1, wherein the building block metadata module creates a data object for data to be written to, the data object being created in response to a data request by a particular file system client within the same building block as the building block metadata module.

3. The computing cluster of claim 2, wherein the particular file system client writes data to the data object by striping the data across at least one storage module in the same building block as the particular file system client and reads the data from the data object by accessing the at least one storage module.

4. A method for reducing inter-switch network traffic in a computing cluster, comprising:

partitioning said cluster into two or more building blocks, each building block including;

a network switch that provides said cluster with access to the building block and is connected to other building blocks using a hierarchy of network switches, at least one computing server, at least one storage server, a storage system;

providing, using a parallel file system, parallel data access to computing servers within each building block and global access to data across said cluster using said switches, said file system including;

a file system client on each computing server serving file system application requesting and having access to data allocated among each of the building blocks, a storage module on each storage server striping data across a storage system within a particular building block the storage module is located and serves data requests for the file system clients in said cluster, and a building block metadata module within each building block tracking a storage location of data allocated by the storage module to the storage system within each building block, and a global metadata module maintaining file system metadata for said file system and said metadata including which building block data is allocated among within said file system; and

scheduling a job of said file system within a building block and spawning said job at a scheduled time;

i) on at least one of at least one computing server in the same building block such that job requirements of the job are met, ii) on at least one of said building blocks where data for said job is located if there is pre-existing data for said job in said cluster, or iii) on at least one of said building blocks having most available resources if there is no pre-existing data for said job in said cluster.

5. The method of claim of 4, further comprising adding at least one additional building block to said cluster if resource utilization of said exceeds a pre-determined threshold.

6. The method of claim 4, further comprising creating, by the building block metadata module, a data object for data to be written to, the data object being created in response to a data request by a particular file system client within the same building block as the building block metadata module.

7. The method of claim 6, further comprising writing data, by the particular file system client, to the data object by striping the data across at least one storage module in the same building block as the particular file system client and reading the data from the data object by accessing the at least one storage module.

8. A computer program product for reducing inter-switch network traffic in a computing cluster, said program product comprising a non-transitory computer readable storage medium having computer readable program code embodied therewith, said program code being executable by a computer to:

partition said cluster into two or more building blocks, each building block including;

a network switch that provides said cluster with access to the building block and is connected to other building blocks using a hierarchy of network switches, at least one computing server, at least one storage server, a storage system; and

provide, using a parallel file system, parallel data access to computing servers within each building block and global access to data across said cluster using said switches, said file system including;

a file system client on each computing server that serves file system application requests and has access to data allocated among each of the building blocks, a storage module on each storage server that stripes data across a storage system within a particular building block the storage module is located and serves data requests for the file system clients in said cluster, and a building block metadata module within each building block that tracks a storage location of data allocated by the storage module to the storage system within each building block, and a global metadata module that maintains file system metadata for said file system and said metadata includes which building block data is allocated among within said file system; and

schedule a job of said file system within a building block and spawn said job at a scheduled time;

i) on at least one of at least one computing server in the same building block such that job requirements of the job are met, ii) on at least one of said building blocks where data for said job is located if there is pre-existing data for said job in said cluster, or iii) on at least one of said building blocks having most available resources if there is no pre-existing data for said job in said cluster.

9. The computer program product of claim 8, said program code being further executable by the computer to create, by the building block metadata module, a data object for data to be written to, the data object being created in response to a data request by a particular file system client within the same building block as the building block metadata module.

10. The computer program product of claim 9, said program code being further executable by the computer to write data to the data object by striping the data across at least one storage module in the same building block as the particular file system client and read the data from the data object by accessing the at least one storage module.