Skip to contentSkip to conversation

Questions regarding the usage of SCSI layout for NFSD PNFS

DISCUSSION Zhou Jifeng 2 participants 2 messages Jan 12 – Jan 14

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Recently, I read the Linux nfs and nfsd codes. The nfs client supports various types of Block type such as PNFS_BLOCK_VOLUME_SIMPLE, PNFS_BLOCK_VOLUME_SLICE, PNFS_BLOCK_VOLUME_CONCAT, and PNFS_BLOCK_VOLUME_STRIPE. However, the nfsd code only supports the SIMPLE/SCSI types. From the perspective of the code, I understand that since Linux nfsd only supports the SIMPLE type, when using it, a SCSI LUN can only be mounted in one folder for use, and it is impossible to achieve the scenario where the same folder cannot simultaneously be associated with multiple physical LUN devices. I wonder if my understanding is correct?

Best regards

#1 lore raw

Hi Zhou,

Quoted 9 lines · Zhou Jifeng

Recently, I read the Linux nfs and nfsd codes. The nfs client supports various types of Block type such as PNFS_BLOCK_VOLUME_SIMPLE, PNFS_BLOCK_VOLUME_SLICE, PNFS_BLOCK_VOLUME_CONCAT, and PNFS_BLOCK_VOLUME_STRIPE. However, the nfsd code only supports the SIMPLE/SCSI types. From the perspective of the code, I understand that since Linux nfsd only supports the SIMPLE type, when using it, a SCSI LUN can only be mounted in one folder for use, and it is impossible to achieve the scenario where the same folder cannot simultaneously be associated with multiple physical LUN devices. I wonder if my understanding is correct?

That is correct.

Implementing mirroring or striping directly in the nfs/blocklayout code would be fairly simple, so if you have a use case feel free to work on it.