Slightly OT: Web server q's - SCSI vs. IDE

[Nik Clayton]

--NklN7DEeGtkPCoo3
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Tue, Oct 30, 2001 at 05:12:00PM -0000, Mark Hughes wrote:
> On Tue, Oct 30, 2001 at 01:37:44PM -0000, Mark Hughes wrote:
> >> Hi guys. I know this is somewhat off-topic, but I know you're a
> >> knowledgable bunch of guys and hope you'll be able to give me some
> valuable
> >> opinions on this subject.
> >
> >> I'm still looking into our new server, and was wondering what peoples
> >> opinions were on SCSI vs. IDE vs. bucket loads of RAM for a
> database-heavy
> >> web site running under FreeBSD....
>=20
> > Bucketloads of RAM :-)
>=20
> Yeah? cool... that should help then. I think the fun bit is going to be
> tuning freebsd, apache, php and mysql to work to their maximum efficiency
> on the hardware we end up with, whatever it is. I'm quite worried that
> we'll get this hardware and not know properly how to set up FreeBSD to get
> the absolute best out of it.
>=20
> Currently we're doing about 40-55,000 hits per day on a vbulletin php/mys=
ql
> forum - and our cobolt raq is very nearly on the verge of not being able =
to
> cope (we had the five minute load average up to 32 on it last night :+) y=
ou
> could almost hear it scream - I was quite impressed that it didn't crash
> though, it normally does when it gets that high).

I set up forums.multiplay.co.uk last Friday.  It's doing about 35,000
hits a day at the moment.  Load average hovers around .1.  That's also
vBulletin/PHP4/MySQL.

> >> From what I've heard in the past, the bottleneck with databases (we use
> >> MySQL) is almost always the hard drive, so to that end, in terms of
> >> performance, how much difference will having SCSI drives make over IDE?
>=20
> > Depends how frequently you're hitting the disk, what the access patterns
> > are like, and how you've layed out the filesystem.  RAID of some sort
> > also makes a difference, as does the disks ability to do tagged queuing.
>=20
> right. What's tagged queueing exactly? googling around a bit, I think I g=
et
> it, but is it a software or a hardware thing?

Hardware.  It's where the OS can send multiple writes to the disk, and
the disk controller reorders them so that the hit the disk in the
optimum order.  There's a discussion about it in sa(4).  It used to be
SCSI only, but some high end IDE drives support it as well.

> >> Also, would having shedloads of RAM (looking at 1GB initially, because
> of
> >> the stupid prices at the moment) make any difference to the IDE vs SCSI
> >> equation? The entire of our database currently is only 100MB odd, is
> there
> >> any way to make MySQL cache *all* that data into RAM?
>=20
> > FreeBSD will cache it on demand.
>=20
> okay... Is there any way of forcing write-caching through the RAM? if the
> system crashes or looses power, a minutes worth of forum posts isn't real=
ly
> going to matter, so is there any way, at the OS level, to force DB writes
> to be cached?

The problem is that you risk losing much more if you do this at the
filesystem level, which is where the OS can do it.

> I know softupdates has this effect with metadata - does that affect all
> writes or just those creating/deleting/moving files?

All writes.

> >> Likewise, would a vinum software RAID-0 or RAID-1 of two IDE drives ma=
ke
> >> them better than a single SCSI drive? How would two SCSI drives in a
> RAID
> >> fair?
>=20
> > It really does depend on your access policy.  You make finger-in-the-air
> > estimates (as I'm doing now), but you really need to benchmark to get
> > the numbers and work out where the bottlenecks are.  vmstat, systat,
> > iostat, and others are all useful in gathering the sort of data you
> > need.
>=20
> unfortunately, being as how we are currently on this evil RaQ server, it
> doesn't have these - it's got vmstat, actually. Any tips for logging the
> output of vmstat in some meaningful way over time?

I don't know of anything specific for the RAQs.

N
--=20
FreeBSD: The Power to Serve             http://www.freebsd.org/
FreeBSD Documentation Project           http://www.freebsd.org/docproj/

          --- 15B8 3FFC DDB4 34B0 AA5F  94B7 93A8 0764 2C37 E375 ---

--NklN7DEeGtkPCoo3
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (FreeBSD)
Comment: For info see http://www.gnupg.org

iEYEARECAAYFAjve80oACgkQk6gHZCw343WuTgCfVtSqveHOwXwMxM8BkflaJj/g
1esAnjxifuRDc9iycsb637O1O4UwVykK
=QiF7
-----END PGP SIGNATURE-----

--NklN7DEeGtkPCoo3--