[K12OSN] IDE if /home is NFS?

Eric Harrison eharrison at mail.mesd.k12.or.us
Thu Apr 1 06:59:04 UTC 2004


On Thu, 1 Apr 2004, Sudev Barar wrote:

>On Wed, 2004-03-31 at 21:50, pnelson wrote:
>> On Wed, 2004-03-31 at 07:12, Dan Young wrote:
>> > On Wed, 2004-03-31 at 06:59, Shawn Powers wrote:
>> > > Am I correct in thinking that as long as the server dishing out /home is 
>> > > scsi with raid5, etc --  the LTSP servers themselves don't need to have 
>> > > SCSI?  I have a really nice server for /home (scsi 320 in raid5 array, 
>> > > gig eth) but hope to have many "mini" labs that all mount /home via NFS.
>> > > 
>> > > I am thinking that my concerns should be about CPU on the LTSP server, 
>> > > and IDE for drives will be fine.  Can anyone confirm this?
>> > 
>> > I believe this is Paul Nelson's setup at Riverdale.
>> > 
>> > -Dan Young
>> > -Parkrose School District
>> > 
>> Yup... We have 4 application servers, dual Xeons, running K12LTSP 4.0.1
>> with IDE drives. Each server NFS mounts /home. Once Mozilla and
>> OpenOffice is up and running, there's not much left for the drives to
>> do. These machines are used by about 115 clients.
>
>Sorry for my limited knowledge but a question here:
>If /home is NFS mounted the advantage (and speed) of having files and
>processing capabilities on the same server (read same bus) goes away and
>network traffic increases? So if users starts OO and the file is mounted
>from NFS then the additional time for file transfer will slow down
>things a bit....Please what is the PCI bus speed in comparison to NFS
>(100mbps or maybe newer 1000mbps)? Yes I can appreciate advantages of
>central file server but what is speed trade off?

Paul has 1000mbps to his servers. But bandwidth is not the main issue
in SCSI vs IDE, it is latency. Under heavy & random I/O, SCSI+NFS will 
still have significantly lower latency than local IDE.

SCSI drives are "smart" about handling heavy random I/O, they will 
survive random-access patterns that will thrash a "dumb" IDE drive to
a halt. LTSP servers often produce the types of random I/O patterns
that are likely to cause IDE drives to fail.

>From the reports I've seen so far, Serial ATA is much better at avoiding
thrash-to-death scenarios, but it is still quite possible to put SATA
under loads that it can't handle but SCSI drives can.

I still highly recommend that K12LTSP servers use SCSI for /home.

-Eric





More information about the K12OSN mailing list