Performance Tuning Recommendations for a RHEL 4 NFS Fileserver
mckinnon at caltech.edu
Fri Dec 1 11:25:53 CST 2006
Be sure the clients are running over TCP and up the rsize and wsize to
something like 32K (client side). It's a two way street... RHEL4WS
should do all of this by default.
Stephen John Smoogen wrote:
> On 12/1/06, Mansell, Gary <Gary.Mansell at ricardo.com> wrote:
>> Dear all,
>> I am about to implement a Dell PE2950 with 4GB RAM attached to an MD1000
>> array with a 15 300GB SAS Drives in a RAID 5 config via a PERC5/e card.
>> The machine is to act as an NFS fileserver for about 100 UNIX and Linux
>> machines and also as a Samba Server for about 100 PC's.
>> I would be interested to hear any recommendations as to tuning
>> parameters that I should incorporate into the configuration.
>> I expect to increase the number of nfsd's from the default 8 to 16 (will
>> this be enough?), will mount with noatime, will tune2fs the filesystem
>> so that it does not get automatically fscked at boot.
>> Are there any other NFS tuning ideas, IO subsystem tuning and Network
>> Tuning Parameters that I should consider?
> NOTE: I have not tested this myself.
> >From what I can tell on other Linux mailling lists, you will want to
> avoid LVM/LVM2 as a subfilesystem on the partition you are exporting.
> Instead use a direct ext3 partition on this hardware. Also making
> sure that the RAID cache is set appropriately for async/sync helps
> greatly also.
More information about the Linux-PowerEdge