From: Adam Flinton 

Geo. wrote:
> "Adam Flinton"  wrote in message
> news:3ebe2445{at}w3.nls.net...
>
>
>>Maybe ext3 was not appropriate either?
>
>
> I can't agree with that, who would setup a file server using redhat and
> change the default file system? Is that real common, does RH suggest that
> somewhere for file server config suggestions?
>

IMHO yes it's common. I have seen a number of Linux servers in production
& they all used a JFS of which RFS & XFS seem in my non scientific
sample to be the most common. OK only a % of those are/were RH. Suse is
quite common over here & I am not sure what the Suse default is (I
think it may be RFS). Most of the people looking after Linux boxes tend to
be ex-Unix admins so they might be going for JFS or XFS because of prior
experience.

> I don't think ext3 was any more inappropriate than NTFS, both are suitable
> for file systems for a file server. It's the RAID0 setup that is
> inappropriate for a data repository.
>

Possibly. Possibly not. Like I say I have this memory being of the recomend
being towards XFS but that might be coz it was ex-Unix conservative types
who prefer tried & test FS'es.

> The correct way to do these benchmarks is pick two guys, one who knows NT
> and one who knows RH, give them a lab used to bug test software (lots of
> different machines with different hardware) and say go to it. Rate each
> machine with each OS and see how the overall results come out.
>
> The reality is in a setup like that, it's not possible to hide the flaws as
> that's what a test lab is designed to expose.
>

Yup.

In fact I would go slightly further wrt SAMBA. I would get a number of
exactly the same machines & give one to MS people, one to RH, one to
Suse etc & possibly one to a bunch from a BSD. Give em a simple
"make it go as fast as possible wrt samba, the choices are yours"
(e.g. wrt fs etc) & then chart the results. If there are any commercial
x86 Unix'es who want in then let them have a box too (e.g. SCO or Solaris
on x86).

If it runs on all of them pace Win2K3 then let's see coz given the samba
code should be the same on all the *ix'es then you'd be able to get an
average for the SAMBA systems (vs Win2K3) as well as a per OS/distrib view.

That would be usefull for all who are considering serving files using an
X86 box & not just derided as marketing fluff.

Adam

--- BBBS/NT v4.01 Flag-4