Date: Thu, 29 Mar 2001 07:52:37 +0200 From: Jochen Kaiser <Jochen.Kaiser@rrze.uni-erlangen.de> To: freebsd-hackers@FreeBSD.ORG Subject: Network Throughput;SMP Message-ID: <20010329075237.A25652@devil.rrze.uni-erlangen.de>
next in thread | raw e-mail | index | archive | help
Hello,
I am doing some tests with a traffic generator atm. It's a SmartBits 6000
and I tried to test my FreeBSD box. I did a test series which did start
at low bandwidth (5MBit) and until 85Mbit all seemed ok. But then I tried
to find the frontiers - and I did.
Test: 93MBit Constant Stream, UDP, PktSize:491 Bytes
Results:
1 Proc: shows clean linear growth of the delay time over the Pckt#
2 Proc: the line shows increased growth at the beginning which than grows
(SMP) in parallel to the other line. (so it shows higher delays)
This astonished me. But netstat -w1 shows the same:
1 Proc shows:
[...]
27083 0 11793296 27079 0 11793296 0
27072 0 11793296 27048 0 11793296 0
27083 0 11795052 27085 0 11793576 0
SMP shows:
[...]
26299 7 11435222 26462 0 9824525 10
26226 8 11441218 3788 0 9999985 15
26312 7 11453406 47716 0 9998115 0
It seems that SMP has some performance losses compared to single proc.
I think this behaviour is already well known. Is there an explanation
for it (maybe SMP overhead, APIC routes IRQ wrong)?
Is there an Analysis?
Used Hardware, Tyan Board Dual PIII 800MHz, BX440, 512MB, 2 fxp, 1xl
Thanks for your time & cheers,
Jochen
p.s: don't stone when this belongs to the smp list, please, just
reroute me :)
--
Jochen Kaiser kind@IRCNET, phone +49 9131 85-28134
Network Administration mailto:jochen.kaiser@rrze.uni-erlangen.de
Regionales Rechenzentrum Universitaet Erlangen-Nuernberg, Germany
GPG public key: http://www.uni-erlangen.de/~unrza2/public_key.txt
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?20010329075237.A25652>