Bike & Snow
2009-May-12 16:59 UTC
[Gluster-users] Exported NFS - written files are zero size
Hello Currently using Gluster 2.0 (final, not release canditate). I've exported a GlusterFS directory by NFS from a Gluster client. I've then mounted this from a couple of NFS clients. The clients are able to create directories on the NFS mounts. However, all files created by the clients have zero filesize and no contents. The clients will read existing files on the NFS GlusterFS OK though. Is this a known issue? How to fix? My export is as follows: /storage/nfs *(fsid=0,rw,no_root_squash,no_subtree_check) Thanks
Bike & Snow
2009-May-13 08:29 UTC
[Gluster-users] Exported NFS - written files are zero size
I was using kernel 2.6.28 on Ubuntu for the Gluster client, which in turn acts as the NFS server. I've tried kernel 2.6.30rc5 and NFS exports now work OK. I can now write to the exported GlusterFS from an NFS client. However, performance is not that great. If I setup an NFS test using a *physical disk* in the machine, I can obtain NFS *write* performance on a network client of 117.9 MB per second. Using a GlusterFS as the exported filesystem, I get these results from the NFS client: Read: 109.3 MB per second Write: 40.6 MB per second So read performance is good but write performance is much lower compared to exporting a physical disk. Around 33% of physical disk. I have tried the write-behind translator on the Gluster client but it makes no difference. Running the disk benchmark directly on the GlusterFS on the Gluster client gives me: Read: 1502.2 MB per second Write: 224.3 MB per second So our write performance on the Gluster client should be more than up to the job of allowing NFS clients to write at much more than 40.6 MB per second. Does anyone have any information on tuning a Gluster client for NFS exports? Thanks