Hello, I have an interesting architecture consisting of a 3 RHEL 5.3 NFS nodes that mounts about 30 TB worth of iscsi disk and presents them as 6 different NFS shares. It is an active-active-active cluster with each node presenting a couple of shares. It works pretty well. I am doubting my decision to use NFS and am wondering if CIFS would be a better route. Here are the NFS stats: getattr lookup access read write readdirplus 20% 16% 7% 44% 9% 1% Each NFS node pushes about 65 MB/s so they are pretty busy. It is a backup/recovery application so I would describe the I/O as lots of small reads/writes. Any thoughts? Thanks, Terry
On Thu, 31 Dec 2009 13:50:10 -0600 Terry <td3201 at gmail.com> wrote:> Hello, > > I have an interesting architecture consisting of a 3 RHEL 5.3 NFS > nodes that mounts about 30 TB worth of iscsi disk and presents them as > 6 different NFS shares. It is an active-active-active cluster with > each node presenting a couple of shares. It works pretty well. I am > doubting my decision to use NFS and am wondering if CIFS would be a > better route. Here are the NFS stats: > > getattr lookup access read write readdirplus > 20% 16% 7% 44% 9% 1% > > Each NFS node pushes about 65 MB/s so they are pretty busy. It is a > backup/recovery application so I would describe the I/O as lots of > small reads/writes. > > Any thoughts? >Why are you looking to switch? As always, the only way to really tell is to bench it out yourself. I can tell you though that the Linux CIFS client doesn't parallelize writes well at all and that will probably hobble your throughput somewhat. -- Jeff Layton <jlayton at samba.org>
Reasonably Related Threads
- dovecot and nfs readdir vs readdirplus operations
- ANNOUNCE: cifs-utils release 5.6 is ready for download
- cifs-utils: regression in (mulituser?) mounting 'CIFS VFS: Send error in SessSetup = -126'
- bug resolve yet for export OCFS2 volume to NFS client ?
- [cifs-utils PATCH v3 0/4] cifs.upcall: allow cifs.upcall to scrape cache location initiating task's environment