Re: SLES 9 SP4 and OCFS2

1) NFS has a good advantage of been state-less - NFS reboot (or even downing 
for a while) should not cause other systems to fail - they just suspend 
until NFS is recovered (of cource, some systems may have a timeout so they 
wil fail after NFS recovers).

2) NFS server MUST have write-back cache (use RAID with battery-backed 
memory) to work fast. It is MANDATORY requirement (through it is not 
official).

3) Don't forget to configure jumbo frames and use them together with NSF.

If you did all 3 things, NFS is pretty fast. I can see a few 
Megabytes/second on Linux-powered NFS and more then 10 - 20 MB/second on 
NetApp/powered NFS.

On the other hand, OCFSv2 have a serious design flaw - any (even short) IO 
problem with the disk system cause all your servers (except the first one) 
to self-fence and reboot in panic. Any network flaw cause the same.

NFS is (usually) slower vs iSCSI, but on 20 - 30% only. If you see more 
difference - check your configuration and your hardware.

----- Original Message ----- 
From: "Werner Flamme" <[email protected]>
To: "suse-oracle" <[email protected]>
Sent: Wednesday, January 16, 2008 1:32 AM
Subject: Re: [suse-oracle] SLES 9 SP4 and OCFS2


Igor Ushkalo   [16.01.2008 09:30]:
> Hi, Werner!
> We're running OEBS and also needed shared FS for few nodes (4-8).
> We tested a lot of different stuff and decided to use NFS.
> As for me, this solution is rock-solid stable and simple to implement,
> in comparison with glutched ocfs and others.
> It could be even fault-tolerant via failofer cluster
> (we're using HP ServiceGuard for this).
>
> Let me know if you need some detailed info.
> Regards,
> Igor.
>
> -- Best regards, Igor Ushkalo (igorus!) - igorus(at)mail.ru, ICQ #19972198
>

Hi Igor,

it's exactly NFS that we use now and what we want to replace. I do not
have a failover cluster for NFS, so any reboot of the NFS host may (or
may not) take down the other systems.

The reason for us to replace NFS is the data throughput. When I use eg.
rsync to copy data from one machine to another (using the IP addresses
that are used to share the NFS), I have a data throughput of about 12
MB/s (same switch). The same data transferred via NFS uses a throughput
of 150 KB/s. It is *very* slow. Copying data to OCFS2 is very fast (~ 80
MB/s), since all systems are connected to the RAID via FC :-)

Regards,
Werner

-- 
Werner Flamme, Abt. WKDV
Helmholtz-Zentrum für Umweltforschung GmbH - UFZ
Permoserstr. 15 - 04318 Leipzig
Tel.: (0341) 235-1921 - Fax (0341) 235-451921
http://www.ufz.de - eMail: [email protected]

-- 
To unsubscribe, email: [email protected]
For additional commands, email: [email protected]
Please see http://www.suse.com/oracle/ before posting



-- 
To unsubscribe, email: [email protected]
For additional commands, email: [email protected]
Please see http://www.suse.com/oracle/ before posting