Well, just on our Network Applicances (NACs) -- which are hardware NFS
servers with RAID5 -- we see *much* higher throughput to disk. AFS of
course minimizes the need for fast reads since most data is cached on the
client, but AFS does little to improve writes, since they still need to
be committed to the fileserver (and they go through the cache!) Writes
to the ODS array *without* AFS seem to be fairly slow, compared to our
NAC and AUSPEX fileservers. (Of course, the ODS solution is *much* cheaper
than an AUSPEX or even a NAC.)
-- Garrett D'Amore.
On Tue, 19 Nov 1996 10:50:04 +0100 (MET) Rainer Toebbicke wrote:
> On Mon, 18 Nov 1996, Garrett D'Amore wrote:
>
> >
> > We don't use SSAs, but we do use ODS with RAID 5. Be warned however that O
DS
> > is slower than snail-snot when it comes to writes. A hardware RAID solutio
n
> > would be preferable, and we are pursuing hardware alternatives at our site.
> >
>
> I cannot fully agree: although we have not done any scientific timing tests,
> the normal 'stress tests' that we ran showed that ODS RAID5 on our 4GB DEC
> (Quantum? Seagate? whatever) disks is not any slower than our IBM 7137
> RAID5, equally with 4GB disks.
>
> A fair comparison is difficult: the IBM is 8 disks on 1 FW diff SCSI bus, the
> Sun one 6 4GB disks on 3 (!) FW single-ended busses. I don't know in how far
> the number of busses plays a role, nor can I remember how much write-back
> cache the IBM RAID actually has. And I do not know what effect the available
> memory (128 MB) on the SPARC 20 server has.
>
> Anyway, this is with standard SCSI-attached disks on the Sun, not yet with
> the StorageArray which has a write-back cache.
>
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
> Rainer Toebbicke http://wwwcn1.cern.ch/~rtb -or- [EMAIL PROTECTED] O__
> European Laboratory for Particle Physics(CERN) - Geneva, Switzerland > |
> Phone: +41 22 767 8985 Fax: +41 22 767 7155 ( )\( )
>
