Re: Postgresql on file system EXT2 or EXT3

Centuries ago, Nostradamus foresaw when andrew(at)libertyrms(dot)info (Andrew Sullivan) would write:
> On Fri, Nov 28, 2003 at 02:46:50PM -0800, Mike Benoit wrote:
>> I would be very interested in seeing actual PGBENCH results with
>> databases on the different file systems, thats the only way you
>> will know for sure which file system is best for the task.
>
> I suspect because of the nature of its workload, in my experience
> pg_bench is lousy for measuring filesystem performance: it always
> bottlenecks somewhere else.
>
> Chris Browne did some work for us some time ago evaluating XFS, JFS,
> and ext3, and concluded that JFS was the best under a high-update
> load; that workload was selected precisely because it was I/O
> bound. I thought he sent the results to the -performance list, but I
> can't put my hands on the email right now. Chris?

Yes, that's right, and the paucity of hard-and-fast details comes from
the fact that the sample workload was, well, pretty proprietary. It
was an honest-to-goodness real workload for one of the registries,
which means that I can't give out copies. (With suitable caveats of
"or else I'd have to kill you," or, more realistically "or else they'd
have to kill me..." :-(.)

The results repeated well, with JFS being ~20% faster than ext3 or
XFS. (I found XFS marginally slower for this benchmark than ext3, but
the difference was small enough that I wouldn't trust that as a True
Conclusion.)

The actual measurements are probably in the internal Systems archives;
I am generally disinclined to give out numbers publicly, in view of
the public unavailability of the workload.

What I had previously reported was actually on the pgsql-admin list...
<http://archives.postgresql.org/pgsql-admin/2003-09/msg00284.php>
--
let name="aa454" and tld="freenet.carleton.ca" in String.concat "@" [name;tld];;
http://cbbrowne.com/info/sap.html
REALITY is a policy phased out early in the Eisenhower administration.