Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 20 Jun 2001 05:28:44 -0700
From:      "Gilbert Gong" <ggong@cal.alumni.berkeley.edu>
To:        <freebsd-hackers@FreeBSD.ORG>
Cc:        <mwlucas@blackhelicopters.org>
Subject:   submit article to sys admin magazine?
Message-ID:  <009a01c0f984$8c64f600$786b85d1@harvestberkeley.org>

next in thread | raw e-mail | index | archive | help
After reading Michael's post at:
http://daily.daemonnews.org/view_story.php3?story_id=2112
and exchanging a few emails with him, I've became convinced it might be
reasonable to try to write an article and submit it to sys admin magazine,
in re: tuning a freebsd box properly.

The goal would be to demonstrate the effect tuning a box properly has on
performance.  The technique would be to take a stock FreeBSD box, run some
benchmarks on it, then rerun those benchmarks after tuning(7) it properly.

Questions:
1) Is anyone already working on something like this?  If so, perhaps I
should not bother :)
2) Can anyone suggest some benchmarks which would be both somewhat
meaningful and relatively easy to implement/duplicate?  I thought it might
be good to give numbers for some kind of http, email, and disk benchmark.
Looking through the ports, I saw what might look interesting were: httperf,
postal, and postmark.

httperf gives me some strange errors, some of which atleast seem to be
caused by bugs in httperf, ie:
httperf --hog --ser=test-server --wsess=100,50,0 --rate 100 --timeout 5
httperf --hog --timeout=5 --client=0/1 --server=test-server --port=80 --uri=
/ --rate=100 --send-buffer=4096 --recv-buffer=16384 --wsess=100,50,0.000
assertion "obj->ref_count > 0" failed: file "object.c", line 179
Abort (core dumped)
It doesn't core dump every time, only sometimes..
Anyone have any experience with this tool, or have a better one to suggest?

postal is also an interesting program, but it doesn't seem to generate
statistics.  It just basically throws as much email at a server as it can.
Which makes it a little harder to use as a benchmark.  What seems the most
obvious thing to do to get around that is to either write a new mail
benchmark, or run the program for a certain length of time, and then
postprocess the mail log on the server machine to see how many mails were
actually delivered in what amount of time.

Finally, I haven't tried postmark yet, and I remember many of the caveats
that were mentioned about it on the lists the past few weeks (or however
long it's been) but I thought it might be interesting because it was in that
other article.

This is my setup so far:
test-server:
P3 600
256MB ram (I can add more, I've got about 768MB sitting around to split
between the server and client test boxes)
Supermicro P6SBU motherboard
onboard aic7890/91 Ultra2 SCSI adapter
IBM DNES-309170W SA30 (8gb scsi hd, at 80MB/s, tagged queueing enabled)
3com 3c905b-tx

test-client1:
P3 400
192MB ram
IBM-DTTA-371440 (ide hd at UDMA33)
kingston 21143-based nic

test-client2:
Celeron 400
128MB ram
Maxtor 90645D3 (ide hd at UDMA33)
onboard intel pro 100+

I can probably get a hold of more client type machines too if it's needed.
I would really appreciate any comments, suggestions, or feedback.  I will
even take kindly to well-meaning flames.
Thanks,
Gilbert


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?009a01c0f984$8c64f600$786b85d1>