Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 02 Mar 2000 19:54:23 +0000
From:      Harry Newton <harry_newton@telinco.co.uk>
To:        freebsd-questions@freebsd.org
Subject:   Help wanted on interpreting benchmarks
Message-ID:  <200003021954.TAA14371@chimaera.locus>

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

Hi,

I have just changed from Debian 2.1 Linux to FreeBSD 3.3-R, running on
Pentium 133 with 64M and an ATA hard drive (UDMA compatible). I
decided to benchmark the two systems and should like any comments on
the results I obtained. Of course, this is not a life-or-death matter
as my main reason for changing over to BSD was the (hopefully) greater
stability (1). I was just a little curious over the tests.

I used unixbench-4.0.1. This runs various tests on the machine and
computes a final merit figure, which of course has no meaning. Of more
interest are the individual test results.

First the file I/O tests: these use the kernel calls write and read.

File I/O 1024 bufsize 2000 maxblocks

kB/sec             Read     Write     Copy
Linux 2.0.36       32525.0  36504.0   19071.0
Linux 2.2.13       31682.0  41044.0   19727.0
Vanilla BSD        19576.0   4088.0    3212.0
BSD + DMA          20476.0   6666.0    6110.0
BSD + DMA + async  20165.0   7200.0    6249.0
BSD + softupdates  19609.0   5955.0    6405.0

File I/O 4096 bufsize 8000 maxblocks

kB/sec             Read     Write     Copy
Linux 2.0.36       36059.0  10737.0    5562.0
Linux 2.2.13       34869.0  43198.0   21629.0
Vanilla BSD        31994.0   4800.0    3987.0
BSD + DMA          32488.0   7200.0    6952.0
BSD + DMA + async  32469.0   7200.0    7003.0
BSD + softupdates  32481.0   7021.0    7136.0

Things I understand are: read rate is similar, with some dependence on
buffer size.

What I don't understand fully: why async and softupdates don't appear
to make any difference in the write and copy tests. The only thing
that appears that to make any difference is switching on DMA. I'm not
sure whether anything is being measured here: for example, the Linux
2.2.13 write figure is somewhere near 43 Meg/s, which must be timing
the transfer to the disk buffer. But this must reflect what actually
happens when you use the system on an application level: so why is BSD
much slower ? What advantage am I accruing with BSD ?

Pipe throughput test

 ( number of iterations per sec whilst pipe is active )
Linux 2.0.36        70621.2
Linux 2.2.13       110108.5
Vanilla BSD         59813.8
BSD + DMA           54026.7
BSD + DMA + async   56167.0
BSD + softupdates   60054.5

Pipe-based Context Switching

Linux 2.0.36        30610.1
Linux 2.2.13        32276.8
Vanilla BSD         14988.1
BSD + DMA           14259.8
BSD + DMA + async   13946.7
BSD + softupdates   14599.1


Any ideas why the BSD should be so sluggish ?

Regards, Harry






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




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200003021954.TAA14371>