[Ocfs-users] ocfs performance

[Wim Coekaerts]

we have benchmarks that show the perfomranec of ocfs for database io
that show clearly that we are at the same speed as raw access.
this is not a comparison done by using dd, but by running oracle
benchmarks.

in fact, even tho ext3 for instance is faster becuase it can do buffered
reads and cheat if you have lots of free memory, it doesn't scale up and
at some point levels off and  raw/ocfs gain rappidly.

I will post this on the website some time as well, the performance group
is polishing the report. 

1.0.2 vs 1.0.9 is a huge difference, io definitely has improved, quite a
bit but not to the pointwhere 1.0.2 was "slow" and this is "fast" even
1.0.2 had decent pure db io. (again, doing a dd to check performance for
instance will not show that type of stuff)

you should probabyl also use e.24 or e.25 of as2.1 as it fixes a number
of vm problems with filesystems, ocfs just ran into it faster since we
have a smaller blocksize (512 vs 4k) so some of the cp problems where
the system would hang after a period of time are actual vm bugs

in general you want to use cp --o_direct etc, 
we just recently fixed a multinode locking problem that could cause
nother hang, and we fixed a mv bug where mv would start doing a cp
insted of just rename


Wim

On Thu, Jul 10, 2003 at 09:02:22AM -0600, Susan M. Wilson wrote:
> Hi,
> 
> I'm new to the list, so forgive me if I'm asking several naive questions 
> that have been answered a dozen times before. 
> I have RHAS with kernel version 2.4.9-e.10smp, running 
> ocfs-support-1.0-2, ocfs-tools-1.0-2, ocfs-2.4.9-e.10-smp-1.0-2.  I have 
> Oracle 9iR2 RAC up and running on an IBM fiber channel architecture.  I 
> find it to be extremely slow compared to other stand alone Oracle 
> databases and believe it to be related to IO off the fiber.  Truthfully, 
> I really don't know whether it's the fiber itself or the ocfs, but I 
> don't believe it's the RAC aspect of things.  In general I have not seen 
> release notes for successive versions of ocfs from 1.0-2 and up and have 
> a couple of questions: 
> Can I expect a performance improvement going to 1.0-9 from 1.0-2?
> Since I have seen no release notes, can anyone tell me some compelling 
> reasons for upgrading.  For instance:
> Does anyone know when/if the problem of crashing server when moving 
> files around on the ocfs will be/has been fixed?
> 
> Thanks,
> Susan
> 
> -- 
> Susan M. Wilson
> Sr. Database Administrator
> The University of New Mexico
> High Performance Computing, Engineering and Research Center
> 
> -- 
> Susan M. Wilson
> Sr. Database Administrator
> The University of New Mexico
> High Performance Computing, Engineering and Research Center
> 
> Desk:  (505) 277-2051
> Cell:  (505) 280-9158
> 
> 
> _______________________________________________
> Ocfs-users mailing list
> Ocfs-users at oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs-users