Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!simtel!news.sprintlink.net!howland.reston.ans.net!EU.net!Austria.EU.net!news.aut.alcatel.at!atusc75!ladavac
From: ladavac@aut.alcatel.at (Marino Ladavac)
Newsgroups: comp.periphs.scsi,comp.os.linux.hardware,comp.os.linux.setup,comp.unix.bsd.freebsd.misc
Subject: Re: SCSI diskIO
Date: 5 Sep 1995 12:17:17 GMT
Organization: Alcatel Austria AG
Lines: 52
Message-ID: <42hf4d$eca@atusks02.aut.alcatel.at>
References: <417ee7$1kp@trauma.rn.com> <DDs7pM.KLA@ritz.mordor.com> <41g8re$11e@news.cloud9.net> <41hqvn$a1p@bonnie.tcd-dresden.de> <41jl81$3d8@news.cloud9.net> <42fs0r$j3s@newsserv.zdv.uni-tuebingen.de>
NNTP-Posting-Host: atusc75.aut.alcatel.at
X-Newsreader: TIN [version 1.2 PL2]
Xref: euryale.cc.adfa.oz.au comp.periphs.scsi:36657 comp.os.linux.hardware:14767 comp.os.linux.setup:18971 comp.unix.bsd.freebsd.misc:5448
Will (zxmgv07@hp12.zdv.uni-tuebingen.de) wrote:
: Hi,
: I tried Bonnie benchmarks on my machine desaster, and found someone else
: who did this with exactly the same SCSI Drive ST31200, but using NCR
: controller instead of the DPT PM2024/A on FreeBSD. The NCR is supposed
: to be slower rather than faster.
: The results are shocking - FreeBSD is way faster in every field except single
: character operation. It does use more CPU time, too ?
: -------Sequential Output-------- ---Sequential Input-- --Random--
: -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
:Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
:desaster 100 1398 39.9 1599 14.4 605 9.7 1368 46.0 1816 19.9 42.2 3.8
^^^^^^^^^ *
:ST31200 100 1353 99.0 3675 48.8 1454 22.6 1377 98.4 4007 37.4 63.9 7.4
:Atls2G 100 1372 97.7 5856 91.4 2420 53.8 1434 98.5 6152 82.8 85.9 15.8
:P90 100 3077 96.9 6193 56.2 2445 31.7 3374 96.7 6377 52.4 78.0 7.1
:desaster Pentium/90,Seagate ST31200N, 1GB, 5400RPM, Linux 1.3.22, DPT PM2024A
:ST31200 486DX2/66, Seagate ST31200N, 1GB, 5400RPM, FreeBSD-2.0, NCR
:Atls2G 486DX2/66, Quantum Atlas 2150, 2GB, 7200RPM, FreeBSD-2.0, NCR
:P90 Pentium/90, Quantum Atlas 1070, 1GB, 7200RPM, FreeBSD-2.0, NCR
:Why is Linux 1.3.22 so slow compared to FreeBSD?
See underlined part.
This is a dead giveaway. It would seem that the underlying kernel support
on Linux is so slow that the machine is I/O limited on Single Char output.
Normally, it should be all related to the CPU overhead on user-to-kernel-to-
user transitions. (I think Bonnie does not do buffering in user space in
Single Character mode.)
FreeBSD si CPU limited in Single Char mode. See the P90 line (same CPU as
desaster's.) desaster is, fairly obviously, I/O limited even is char mode.
Otherwise, FreeBSD CPU percentages scale as one would expect: roughly twice
the load for twice the performance. (as long as the system is I/O limited)
:Cheers, Michael Will
No Problem(tm)
/Alby
--
._______V_______. Dipl.-Ing. Marino Ladavac (marino.ladavac@aut.alcatel.at)
| A L C A T E L | Systems Engineer Network Management Applications
+---------------+ Alcatel Austria AG. Scheydgasse 41. 1211 Wien. Austria
Tel:(+43 1) 27722 3769 Fax:(+43 1) 27722 171