Exporting a GPFS file system using NFS
- Create and mount the GPFS file system. In the examples, we
assume a file system with a local mount point of /gpfs.
For performance reasons, some NFS implementations cache file information on the client. Some of the information (for example, file state information such as file size and timestamp) is not kept up-to-date in this cache. The client may view stale inode data (on ls -l, for example) if exporting a GPFS file system with NFS.
If this is not acceptable for a given installation, caching can be turned off by mounting the file system on the client using the appropriate operating system mount option (for example, -o noac on Linux® NFS clients). Turning off NFS caching results in extra file systems operations to GPFS, and negatively affect its performance.
Note:- Ensure that all GPFS file systems that you use to export data via NFS are mounted with the syncnfs option to prevent clients from running into data integrity issues during failover. Prior to mounting a GPFS system, it is a good practice to run the mmchfs command to set the syncnfs option, -o syncnfs.
- Ensure that NFS clients mount with the -o hard option to prevent any application failures during network failures or node failovers.
- If caching is turned on for the NFS clients, files that are migrated to the cloud storage tier by using Transparent cloud tiering remain in the co-resident status, and the capacity is not freed from the file system. However, if caching is disabled, the files are moved to the non-resident status and the capacity is freed. In this case, there is a negative impact on the performance. Therefore, there is a tradeoff between capacity and performance, and administrators must take a judicious decision depending on the business requirements.
- Make sure that the clocks of all nodes in the GPFS cluster are synchronized. If this is not
done, NFS access to the data, as well as other GPFS file system operations, may be disrupted.
NFS relies on metadata timestamps to validate the local operating system cache. If the same directory is either NFS-exported from more than one node, or is accessed with both the NFS and GPFS mount point, it is critical that clocks on all nodes that access the file system (GPFS nodes and NFS clients) are constantly synchronized using appropriate software (for example, NTP). Failure to do so may result in stale information seen on the NFS clients.
- Ensure that NFS is properly configured and running.
For Linux nodes, information on configuring NFS can be obtained at the LinuxDocs.org.
For AIX® nodes, refer to AIX in IBM® Documentation for information about configuring NFS.