Being efficient with IO

[biatche]

I need some suggestions on how to maximize I/O for a public iSCSI on ZFS system.

I'll be booting Windows clients off iPXE - iSCSI, which will be on a zvol. Each client will have its own ZFS clone, after which, when a session ends, the clone will be destroyed. Since the writes during this session aren't important, is there a way to redirect the writes to another physical disk? If there are a lot of writes to the main drives, wouldn't this impact read speed? Fast reads are more important in my case here than writes (since they will be discarded).

Is there a better way?

 

[vermaden]

@biatche,

Make the recordsize of these iSCSI ZVOLs the same as the filesystem block size (in that case probably NTFS).

% zfs get -r recordsize YOURPOOL

 

[biatche]

That's a good optimization to note of. But still, writes will occupy I/O unnecessarily to the main drives

 

[SirDice]

How about adding a ZIL? Another option may be to use mount_unionfs(8).

 

[biatche]

How about adding a ZIL? Another option may be to use mount_unionfs(8).

ZIL is nice, except it still has to write eventually. I've considered UnionFS, but how will this work with NTFS? UnionFS is indeed ideal, but I can't figure out how to overlay NTFS with UnionFS.

 

[SirDice]

I've considered UnionFS, but how will this work with NTFS? UnionFS is indeed ideal, but I can't figure out how to overlay NTFS with UnionFS.

Ah, I think I missed that part. I don't think it's possible to use UnionFS in that case.