> > For single-user tasks you will probably be better off by
> > getting a gray box with Athlon 2600+ with 3 Gigs of memory and
>
> > IDE disks and running Linux or *BSD .
>
> I'm not particular to
> Macs, but I've got to say, that stock out the box, postgres
> loves it. That old G3 was faster than the Sun, and still is
> faster than my (years newer) linux laptop (on which I've done no
> performance tweaking). So maybe a dual G4 Xserver would scream?
>
> Any suggestions? It's still not too late for us to change our
> minds on this one.
I can't recommend macs for either brute force speed or price/performance.
My current flock of machines are mostly OSX g4 boxes (single 400s and dual 800), with a couple of linux boxen thrown in
forgood measure.
The mac's biggest issues are:
1) Tweakability - you've got one file system, and it doesn't really do useful mount options like noatime.
2) There are bugs in mount and traversing symlinks that make it hard to move pg_xlog onto another file system and
retainperformance (in 10.1.5, I don't have test hardware with enough drives to test 10.2)
3) vm_stat gives vm status, iostat gives nothing. Looks like this is working on my 10.2 laptop, but it's annoyed me for
awhile on 10.1.5
4) SW raid is not that much of a help for speed.
5) Bus bandwidth is a good factor behind x86 linux boxen. (DDR ram isn't really taken advantage of in current designs)
Having said that, I'm getting reasonable performance out of all the macs, in fact, I'm getting reasonably similar
performanceout of all of them desplite the 4x difference in processor power. And that's because they basically have
thesame low end disk system.
I'm piecing together something that I hope will be faster out of a x86 box loaded with more drives in a sw raid
mirroringsetup. (tentative is 1x system+logs, mirror for pg_xlog, mirror for pg data)
I'm planning on running some comparative benchmarks prior to going live, so I should be able to tell how much faster it
is.
eric