Return to BSD News archive
#! rnews 5430 sserve.cc.adfa.oz.au
Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.ecn.uoknor.edu!constellation!convex!news.netins.net!solaris.cc.vt.edu!hookup!lll-winken.llnl.gov!decwrl!svc.portal.com!news1.best.com!shell1.best.com!not-for-mail
From: dillon@best.com (Matt Dillon)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Re: Was: ISP , now: How Using FBSD?
Date: 31 May 1995 02:36:39 -0700
Organization: Best Internet Communications, Inc. (info@best.com)
Lines: 115
Message-ID: <3qhdbo$4hd@shell1.best.com>
References: <3pqb92$lq2@pt9201.ped.pto.ford.com> <D92JoB.B0I@ritz.mordor.com> <3puu0i$nqi@shell1.best.com> <3q4tnq$loa@arrow.va.pubnix.com>
NNTP-Posting-Host: shell1.best.com
:In article <3q4tnq$loa@arrow.va.pubnix.com>,
:Kurt J. Lidl <lidl@va.pubnix.com> wrote:
:>In article <3puu0i$nqi@shell1.best.com>, Matt Dillon <dillon@best.com> wrote:
:>> I'm in a good position to compare BSDI to FreeBSD as well.. we
:>> started out running BSDI 1.1 but are now beginning to shift
:>> everything over to FreeBSD. Frankly, the only reason we went
:>> with BSDI in the first place was for the technical support...
:>> so much for that, it isn't all that hot, and BSDI tends to hold
:>> their cards close to their chest and penny pinch a lot...
:>
:>I would have to disagree with you on both of those points. First,
:>the service, in my opinion, is worth every penny. When I started
:>testing out the the 2.0 release (during the alpha and beta tests),
:...
hmm... well, everyone has their own opinion. Can't argue with
that. You asked me to post numbers though...
:>
:>> FreeBSD beats the hell out of BSDI when it comes
:>> to SCSI throughput, especially with those cheap $70 NCR PCI SCSI cards.
:>
:>Post numbers, I don't believe it.
BSDI is still using a very, very old SCSI subsystem as far as I
can tell. The best I can get out of our BSDI system on a good day
is on the order of 1.2 MBytes/sec writing and about the same reading.
That's with a PCI Buslogic SCSI card on a 90 MHz pentium machine. The
FreeBSD machine, identical except that it uses an NCR PCI SCSI
card, with similar loading (very low) gets 3.3 MBytes/sec
writing and 6.5 MBytes/sec reading. Both machines with 4G baracuda's,
the testing done on a reasonably unloaded partition.
It's quite consistant.
Here are the FreeBSD timings on a 64MB file (enough to defeat
caching) using, admittedly, a simple dd test:
blob:/u4# time dd if=/dev/zero of=test bs=32k count=2048
2048+0 records in
2048+0 records out
67108864 bytes transferred in 20 secs (3355443 bytes/sec)
0.015u 7.713s 0:20.48 37.6% 63+290k 8+1171io 0pf+0w
blob:/u4# time dd if=/dev/zero of=test bs=32k count=2048
2048+0 records in
2048+0 records out
67108864 bytes transferred in 21 secs (3195660 bytes/sec)
0.070u 8.131s 0:20.95 39.1% 63+289k 9+1194io 0pf+0w
blob:/u4# time dd if=/dev/zero of=test bs=32k count=2048
2048+0 records in
2048+0 records out
67108864 bytes transferred in 20 secs (3355443 bytes/sec)
0.031u 7.807s 0:20.63 37.9% 63+290k 8+1188io 0pf+0w
blob:/u4#
blob:/u4# time dd if=test of=/dev/null bs=32k count=2048
2048+0 records in
2048+0 records out
67108864 bytes transferred in 11 secs (6100805 bytes/sec)
0.037u 5.379s 0:11.33 47.6% 67+306k 1034+12io 0pf+0w
blob:/u4# time dd if=test of=/dev/null bs=32k count=2048
2048+0 records in
2048+0 records out
67108864 bytes transferred in 11 secs (6100805 bytes/sec)
0.023u 5.213s 0:11.08 47.2% 64+290k 1034+12io 0pf+0w
blob:/u4# time dd if=test of=/dev/null bs=32k count=2048
2048+0 records in
2048+0 records out
67108864 bytes transferred in 11 secs (6100805 bytes/sec)
0.022u 5.276s 0:11.03 47.9% 64+295k 1034+11io 0pf+0w
blob:/u4#
The BSDI (1.1) timings on a 32MB file (sorry, too tired to run
it on a 64MB file). I'm thinking the data-read timings might
be off due to the size of the buffer cache, which is pretty
big due to the amount of ram on the system.
shell1:/u5/tmp# time dd if=/dev/zero of=test bs=32k count=1024
1024+0 records in
1024+0 records out
33554432 bytes transferred in 37 secs (906876 bytes/sec)
0.010u 4.830s 0:37.75 12.8% 0+0k 24+4133io 0pf+0w
shell1:/u5/tmp# time dd if=/dev/zero of=test bs=32k count=1024
1024+0 records in
1024+0 records out
33554432 bytes transferred in 32 secs (1048576 bytes/sec)
0.010u 5.380s 0:33.78 15.9% 0+0k 9+4129io 0pf+0w
shell1:/u5/tmp# time dd if=/dev/zero of=test bs=32k count=1024
1024+0 records in
1024+0 records out
33554432 bytes transferred in 34 secs (986895 bytes/sec)
0.010u 5.360s 0:33.90 15.8% 0+0k 11+4129io 0pf+0w
shell1:/u5/tmp#
shell1:/u5/tmp# time dd if=test of=/dev/null bs=32k count=1024
1024+0 records in
1024+0 records out
33554432 bytes transferred in 46 secs (729444 bytes/sec)
0.010u 2.830s 0:45.75 6.2% 0+0k 4100+1io 0pf+0w
shell1:/u5/tmp# time dd if=test of=/dev/null bs=32k count=1024
1024+0 records in
1024+0 records out
33554432 bytes transferred in 57 secs (588674 bytes/sec)
0.040u 2.500s 0:57.80 4.3% 0+0k 4100+3io 0pf+0w
shell1:/u5/tmp# time dd if=test of=/dev/null bs=32k count=1024
1024+0 records in
1024+0 records out
33554432 bytes transferred in 47 secs (713924 bytes/sec)
0.020u 2.540s 0:47.53 5.3% 0+0k 4103+2io 0pf+0w
shell1:/u5/tmp#
-Matt
--
Matthew Dillon VP Engineering, BEST Internet Communications, Inc.
<dillon@best.com>, <dillon@apollo.west.oic.com>
[always include a portion of the original email in any response!]