Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 02 Dec 1996 22:18:47 -0800
From:      Jason Thorpe <thorpej@nas.nasa.gov>
To:        "David S. Miller" <davem@jenolan.rutgers.edu>
Cc:        jkh@time.cdrom.com, dyson@freebsd.org, dennis@etinc.com, kpneal@pobox.com, hackers@freebsd.org, torvalds@cs.helsinki.fi, lm@engr.sgi.com, iain@sbs.de, sparclinux@vger.rutgers.edu
Subject:   Re: TCP/IP bandwidth bragging 
Message-ID:  <199612030618.WAA06227@lestat.nas.nasa.gov>

next in thread | raw e-mail | index | archive | help
On Mon, 2 Dec 1996 21:17:34 -0500 
 "David S. Miller" <davem@jenolan.rutgers.edu> wrote:

 > pinhead.  If you think it does not, why does the government spec
 > lmbench numbers for all purchases these days?  What concrete numbers

We spec NASPAR benchmarks (home-grown, specifically to measure
performance of a system under our typical workload).  I sort of
hold the opinion that "general purpose" benchmarks like lmbench
are fun to play with, but I certainly wouldn't claim one OS is
better than another based on "general purpose" benchmark numbers.

 > is a dinky 40MHz SparcClassic (4k I and D caches, thats it) with 40MB
 > of ram and two SCSI disks.  The load never goes over 4.

Load average measures the average number of runnable processes.  If
everyone's waiting for I/O to complete, well, they're not runnable...

Actually, specifically, load average is:

	load avg = nrunnable + number of processes sleeping in \
	    uninterruptable sleeps for < 1 second.

So, your load average of 4 could be "lots of processes not runnable because
they're blocking on I/O" plus "a few process in uninterruptable disk wait" :-)

So, maybe you load average is yet another example of a meaningless Linux
number :-)  My point is that claiming a load average number doesn't
really tell us anything.  Why don't you tell us exactly what's going
on in the system.  Then we might be impressed (depending on what
you tell us :-).  As it stands, your bragging and number flashing
is, well, just annoying.

IMO, for the kind of work I do (kernel development), a much better
`benchmark' is to run a profiling kernel and then do a "make build"
(which is what NetBSD uses to build releases).  With the profile
data, you can find places where you actually spend a significant
amount of time, and improve them.  Sometimes the profiling results
are surprising.

I see Larry's on the Cc: now (amazing how these explode so we can
all prove our manlyness to an audience :-)... I saw him at SC'96,
and tried to bump into him a couple of times... never got a chance
to hook up with him... I was rather looking forward to arguing
about benchmarks :-)

Jason R. Thorpe                                       thorpej@nas.nasa.gov
NASA Ames Research Center                               Home: 408.866.1912
NAS: M/S 258-6                                          Work: 415.604.0935
Moffett Field, CA 94035                                Pager: 415.428.6939



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199612030618.WAA06227>