Date: Wed, 11 Mar 1998 12:23:13 -0800 (PST) From: Hugh LaMaster <lamaster@george.arc.nasa.gov> To: current@FreeBSD.ORG Cc: lamaster@george.arc.nasa.gov Subject: Re: my worldstone Message-ID: <199803112023.MAA18435@george.arc.nasa.gov> In-Reply-To: <199803111939.NAA27575@home.dragondata.com> from "Kevin Day" at Mar 11, 98 01:39:57 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> > It will be nice if people with very fast systems and i/o subsystems to
> > post their world stone. It has been alluded in the past that we have
> > a bottle neck in the system we seem to hover around 100 minutes. For
BTW, I notice some folks posting "make world" and others "make buildworld".
What are the "official" parameters for "world stone"?
> > instance, Simon Shapiro posted in the past that increasing the i/o
> > subsystem like by using a DPT couple with fast disks didn't seem
> > to improve his relative world stone benchmark.
> >
> yy
> As a point of reference:
>
> My dual PII/333, two 10000 rpm 4G ultra wide scsi drives, each on their own
^^^^^^^
What chipset? I assume LX chipset, because I don't think the Natoma
boards for PII were supposed to go over 300 MHz, but, you could be
overclocking.
How much memory? What memory? SDRAM? w/ ECC? Are the memory timings
set properly for SDRAM in the BIOS? (x-1-1-1 with SDRAM?).
> adaptec controller(/usr/src in one, and obj in the other), 256M Ram, and no
> other tasks running still took around one hour 45 minutes to complete a make
> buildworld -j4.
Curiously, for me this (that is, "make buildworld") was
~ 2 hrs 30 minutes for me on a PPro 200, Natoma chipset
(ASUS P/I-6NP5 I think it was - I don't remember the
exact model number) with 3.0-current, in January.
"make installworld" added another ~1/2 hour. I don't
have the system anymore, or exact timings.
PPro200, Natoma, 64 MB FP parity DRAM with parity enabled
Buslogic BT-958 SCSI controller
Quantum Atlas 4.x GB disk - 1 disk, everything on it, filesystems synchronous
3.0-Current as of mid-January, 1998.
Since this system has pretty bad write bandwidth, I assume that
its relatively good performance on "world stone" is due to the
low-latency L2 cache.
> Is it just memory bandwidth we're hitting? Putting both directories one
> drive slightly slowed things down.... Same with going to a UP kernel. But
> nowhere near the differences it made on a similar dual P/200.
Write memory bandwidth should be significantly better on the PII/333
than PPro200. L1 cache is bigger, faster, L2 cache is bigger, *slower*.
Significantly longer latency. According to the specs.
Could the longer L2 cache latency hurt PII/333 performance that much?
On some other benchmarks, PII/233 - PII/266 on Natoma seems roughly
the somewhere around ~ PPro200. PII/333 on LX "should" be way faster.
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199803112023.MAA18435>