Planning Volume 2, Control Workstation and Software Environment

Planning for General Parallel File System (GPFS)

General Parallel File System for AIX provides concurrent shared access to files spanning multiple disk drives located on multiple nodes. This licensed program provides file system service to parallel and serial applications |in an SP system or an HACMP/ES cluster environment.

|You can use GPFS in several system environments and |configurations. This discussion is limited to using GPFS in a system |configuration where PSSP is running, like one of the following |configurations: |

• |An SP or clustered enterprise server system with an SP switch network and |PSSP with the Virtual Shared Disk and Recoverable Virtual Shared Disk option |installed and running.
• |Any SP system or clustered enterprise server system running PSSP and the |HACMP/ES licensed program. This is considered by GPFS to be an HACMP/ES |cluster environment. |

|In the SP environment, the boundaries of the GPFS cluster depend on |the switch type being used. With the SP Switch2, the GPFS cluster is |equal to all of the nodes in the system. With the SP Switch, the GPFS |cluster is equal to the SP system partition in which it is configured. |Within a GPFS cluster, the nodes are divided into one or more GPFS |nodesets. A nodeset is a group of nodes that all run the |same level of GPFS and operate on the same file system. The nodes in |each nodeset share a set of file systems that are not accessible by the nodes |in any other nodeset.

You can modify your GPFS configuration after it has been set, but your reward for a little consideration before installing it is a more efficient file system.

Hardware and operating environment considerations:

• |GPFS requires the Recoverable Virtual Shared Disk component of |PSSP. If Recoverable Virtual Shared Disk is on multiple nodes within a |system partition, each node must have the same level of PSSP.
• If you are using twin-tailed disks, you must select an alternate node as a backup Virtual Shared Disk server.
• Do you have sufficient disks and adapters to provide the needed storage capacity and required I/O time?
• |GPFS is supported only in systems with the SP Switch or the SP |Switch2.

File size considerations:

• How much data will be stored and how large will your files become?
• How often will the files be accessed?
• Do your applications handle large amounts of data in single read/write operations or is the opposite true?
• How many files do you anticipate handling in the future?

Data recovery considerations:

• Node Failure: GPFS automatically reconfigures itself to continue operations without the failing node.
• Virtual Shared Disk server and disk failure: Your recovery strategy depends on your answer to the question, 'Is your primary concern loss of data, loss of access to data, or do you need protection from both server and disk failure?'
  1. If data loss is your concern, a RAID device might be the best solution.
  2. If data access is your concern, twin-tailed disks could be your solution.
  3. If both data loss and access are potential problems, first consider mirroring at the logical volume manager for data recovery. If mirroring does not fit your system needs, another option is replication, which automatically creates and maintains copies of all file information.
• |Connectivity failure: an adapter failure is treated as a node |failure on SP Switch and one switch plane SP Switch2 systems. On SP |Switch2 systems with two switch planes, a condition with two adapter failures |on the same node is treated as a node failure.

|Details on planning and implementing these strategies and other |methods can be found in the book IBM General Parallel File System for |AIX: Concepts, Planning, and Installation Guide.