|
Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com |
From: Craig Sanders (cas
taz.net.au)Date: Thu Sep 27 2001 - 17:09:03 CDT
On Thu, Sep 27, 2001 at 01:11:34PM +0300, Liviu Daia wrote:
> > > - if possible, use OpenBSD or FreeBSD with softupdates rather than
> > > Linux;
> >
> > alternatively, configure and use linux properly rather than believe
> > propaganda.
>
> Better yet, try both (take your time to tune them), then choose.
yep, can't disagree with that at all. in fact, i agree with it 100%.
> > > - if you insist on using Linux, use a 2.2.x kernel, not a 2.4.x one,
> > > and turn OFF the swap; buy more RAM if you feel you're running out
> > > of it;
> >
> > 2.4.x kernels are perfectly adequate, indeed much better than 2.2.x
> > kernels.
>
> Not in my experience. Of course, YMMV.
i've had no problems with 2.4.x kernels at all, and there have been
significant performance improvements (in nearly all major areas - disk,
network, SMP, etc) over 2.2.x
there have been several variations on paging algorithms in the 2.4.x
kernels so far - but i can't say i've noticed any real difference in
performance with any of them...the differences only appear when memory
is tight. if you tune things (e.g. number of smtp/smtpd processes, etc)
so that you have a reasonable safety margin of free RAM then it's not
going to be a problem.
> > and there's no need to turn off the swap.
>
> If you run into swap, performance would suck no matter what.
true, but it's better to swap and slow down than to run out of VM and crash.
> For some reasons, Linux 2.4.x seems to exacerbate this problem (again,
> in my experience; standard disclaimer apply).
can't say that i've noticed that - but my policy is to have more RAM
than i think i need...RAM is dirt cheap these days, an extra $100 or so
to add another 256MB is nothing compared to the cost of the rest of the
system.
> > my testing a few months ago proved that RAID5 is significantly faster
> > than RAID0+1 for the kind of disk loads you see on mail servers.
>
> That's because:
>
> > the tests were performed using the exact same hardware, and same
> > software, on the same machine - dual P3-866 with 512MB RAM, hardware
> > raid card with 32MB non-volatile cache, and 6 x 18GB IBM drives.
>
> How many of the RAID cards out there have 32+ Mb of non-volatile
> cache, and how much do they cost?
no, because the RAID0+1 tests were performed on the same hardware as the
RAID5 tests - so they both had the benefit of the non-volatile cache.
everything else was equal, the only difference was RAID5 vs RAID0+1.
the results surprised me, too. i expected RAID0+1 to beat RAID5 because
everything i had read on the topic said that it would.
now it may be that the RAID card i used had a crappy algorithm for
RAID0+1 and a good one for RAID5, but i wasn't in any position to test
that. my purpose was to test the hardware i had and decide the best way
to configure it for the task at hand (a new mail server to replace the
old one).
i'd recommend that anyone building an important server do the same thing
with whatever hardware they have bought for the job. it might take an
extra day or three reformatting and rebuilding the system several times
to run tests, but it's worth that time to get it right BEFORE the system
goes into production use.
i do this as a matter of course now - i've had a few incidents
where benchmark results has flatly contradicted what i previously
believed...so now i believe nothing until i've verified it myself.
one of these days i'll benchmark RAID5 vs RAID0+1 on software raid and
try to find out which is better in that situation.
> > > - use ext2; Xfs and Jfs are not yet ready for prime time, and
> > > everything else is slower than ext2 for mail handling;
> >
> > ext2 would be one of the worst choices for a mail server's filesystem.
>
> Not according to some tests I've seen recently. If you're
> manipulating many small files, ext2 performs consistently better than
> Reiserfs (OTOH, ext2 loses big time for large files --- but that's not
> really relevant here).
not in my experience or in my benchmarks. ext2 performance really
suffers when you have lots of little files, whereas reiserfs really
shines with lots of little files.
> > xfs is reliable and ready for prime-time in my experience.
> [...]
>
> On Linux? How nice.
yes, it is rather.
craig
-- craig sanders <castaz.net.au>
Fabricati Diem, PVNC. -- motto of the Ankh-Morpork City Watch - To unsubscribe, send mail to majordomo
postfix.org with content (not subject): unsubscribe postfix-users
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]