Date: Thu, 5 Jun 1997 03:10:10 -0400 (EDT) From: Kenneth Merry <ken@housing1.stucen.gatech.edu> To: smp@csn.net (Steve Passe) Cc: brian@mpress.com, freebsd-smp@FreeBSD.ORG Subject: Re: 2xPP200 vs 2xPP150 Message-ID: <199706050710.DAA09276@housing1.stucen.gatech.edu> In-Reply-To: <199706050140.TAA24867@Ilsa.StevesCafe.com> from Steve Passe at "Jun 4, 97 07:40:39 pm"
next in thread | previous in thread | raw e-mail | index | archive | help
Steve Passe wrote...
> Hi,
>
> > I have access to two systems:
> >
> > And noticed the following:
> >
> > MB CPUs RAM eth hdc hd kbt
> > Asus MB 2xPP200 256MB de0 2940W QUANTUM XP34550W LXQ1 2:48
> >
> > Tyan MB 2xPP150 64MB de0 2940W SEAGATE ST15150W 0009 2:49
> >
> > kbt = kernel build time with 'make -j 4'
> >
> > So what is going on?
>
> be more specific about exactly what 'kbt' includes, ie make depend?
>
> for me if I "make clean && make depend", then "time make -j8"
> runs in 89-95 seconds (1:29/1:35) on an 2xP6-233x512,
> and around 1:45 on a 2xP6-200x512
>
> there should be a bigger difference between your 2 systems, and both seem slow.
> what times do you get if running only one CPU on each?
My system is very similar to one of Brian's:
MB CPUs RAM eth hdc hd kbt
Asus MB 2xPP200 128MB de0 3940UW QUANTUM XP34550W LXY4 1:50
That time is just build time for a generic kernel, done with
make -j 8. (i.e. not including make clean or make depend) My custom
kernel, compiled with make -j 8, takes about 1:45.
If I had to guess at reasons for the slowdown, I'd say try the
following:
use make -j 8 instead of make -j 4
- this still wouldn't explain the lack of difference in times between the
two machines
upgrade the firmware on your drive from LXQ1 to LXY4.
- Justin said a little while back that there are bugs with the older
firware revisions on at least the Atlas II, and I think the Atlas I
that cause tagged command queueing to not work properly. I was able to
successfully upgrade my firmware, and I know that Satoshi has done it
as well. (he has a XP34550W as well)
- Once you get the firmware upgraded, recompile a kernel with these
options:
AHC_TAGENABLE
AHC_SCBPAGING_ENABLE
(I have AHC_ALLOW_MEMIO turned on as well, I'm not sure what that
does.)
It's possible that upgrading the firmware and turning on tagged
comand queueing might fix the performance difference. It depends on how
disk-bound the kernel compile is. One thing I noticed is that it doesn't
look like the kernel Makefile uses -pipe. That would probably speed things
up as well. Well, I just tried it -- using make -j 8, it cut my generic
kernel compile time by 7 seconds to 1:43. Not a huge change, but it's a
little faster.
Of course it could turn out that it's not a drive problem at all.
Also, I'd like to know whether the times above are for a make depend and a
make, or for just a make. Also, which kernel? Since my and Brian's
systems are very similar, the performance numbers should be similar.
(other things that might factor into my performance -- I was running X
during the tests, and I've only got 256K P6-200's, and I've got standard
60ns DRAM)
Ken
--
Kenneth Merry
ken@ulc199.residence.gatech.edu
Disclaimer: I don't speak for GTRI, GT, or Elvis.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199706050710.DAA09276>