Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 03 Mar 2003 17:31:41 -0800
From:      Rick Jones <rick_jones2@hp.com>
To:        ia64@freebsd.org
Subject:   initial netperf tests on rx2600
Message-ID:  <3E6401FD.E2B384E3@hp.com>

next in thread | raw e-mail | index | archive | help
Folks -

Thank's to Marcel's fixing of the EFI trap problem, and his installing
bits for me :) I was able to run a first pass at netperf aggregate
TCP_RR perf under FreeBSD 5.0 on the rx2600

There are something like four OSes one can boot on IPF/Itanium2 at the
moment - HP-UX, Linux, FreeBSD and Windows (OpenVMS is probably not
toooo far behind, but those four are the ones "shipping" today).  I am
able to include figures for three of the four here.

	    Aggregate, single-byte, netperf TCP_RR results
	 Truncated to the Nearest 100 Transactions per Second
		       Using Core GbE (rx2600)
		    Two-CPUs present, One Enabled

  Concurrent                       System
              sut220      sut220     sut220
   Streams    rx2600      rx2600     rx2600
           HP-UX 11.22   RH AW 2.1 FreeBSD 5.0
  --------------------------------------------------------------------
      1 |      10900        7600       6700
      2 |      20900       11900      13500
      4 |      38300       19400      20800
      8 |      41800       32600      27300
     16 |      44800       37800      48300
     32 |      45100       83200      48100
     40 |      45000       83000      48400
     48 |                  80500

The "load generators" here were a pair of HP rp2470's, each with 2, 750
MHz PA-8700 CPUs, and using an add-on GbE NIC based on the tg3 core.

The tests ultimately take the SUT (System Under Test) to 100% CPU
utilization.

All NIC settings are at their defaults.  I know how to tweak coalescing
parms under UX, but not FreeBSD or Linux :)

It seems clear that both Linux and FreeBSD have some sort of heuristic
that kicks-in to alter some settings as load increases.  

happy benchmarking,

rick jones

BTW, what would be the "best" way for netperf to obtain programatic CPU
utilization figures. At present, it measures CPU util by taking "idle
rates" before a test and comparing them with during a test.  What api
might netperf call to retrieve that sort of data?

-- 
Wisdom Teeth are impacted, people are affected by the effects of events.
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to raj in cup.hp.com  but NOT BOTH...

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-ia64" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3E6401FD.E2B384E3>