*BSD News Article 43312


Return to BSD News archive

Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!newshost.marcam.com!zip.eecs.umich.edu!newsxfer.itd.umich.edu!news.itd.umich.edu!pha
From: pha@umich.edu (Paul H. Anderson)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Re: NCR53C815?
Date: 11 Mar 1995 15:55:58 GMT
Organization: University of Michigan
Lines: 52
Message-ID: <3jsh6e$8d3@lastactionhero.rs.itd.umich.edu>
References: <3jikpm$2b8@marton.hsr.no> <3jkjdr$hip@news.rrz.uni-koeln.de>
NNTP-Posting-Host: grenade.ifs.umich.edu

In article <3jkjdr$hip@news.rrz.uni-koeln.de>,
Stefan Esser <se@MI.Uni-Koeln.DE> wrote:
>In article <3jikpm$2b8@marton.hsr.no>, vebjorn@hsr.no (Vebjoern Forsmo) writes:
>|> I've got an Intel NCR 53C815-based PCI SCSI-controller.  Does anyone know 
>|> of a kernel that supports this controller?
>
>The NCR 53c810, 815 and 825 are supported in
>FreeBSD-current (and ought to be in the latest
>SNAP). In fact, the only difference (visible 
>to the driver) between the 810 and 815 is the
>different PCI device ID.
>
>We didn't have the hardware to do WIDE SCSI
>tests with the 53c825, but got success reports.
> Stefan Esser				Internet:	<se@ZPR.Uni-Koeln.DE>

I am running FreeBSD 1.5.1R with Stefan and Wolf's driver on a P-90
with a Tyan MB and Tyan NCR 53C825 controller along with 4 2GB
FW barracuda drives (7200RPM).  I took the driver and patched it
into the kernel, it built with no problems, as I recall.

I have had no problems with the software, but operator error took
a bit of messing with to get straightened out.  You have to use
the -d0 option on newfs, because it defaults to 4 milliseconds (oops).

Performance is good, but possibly less than I was hoping for.  Using
dd to/from the raw devices gives approximately 5MB/sec (large
sequential files, 1K/8K fresh filesystem).  Doing mmap
of large files and touching all the pages gives about 3.0 MB/sec.
Doing simultaneous drives with raw reads results in peak performance of
about 8.0MB/sec for the controller (I think this was with three drives
running flat out).  I suspect the cost of SCSI commands begins to
cost more than actually moving data.  I recall that I did these
tests with a number of different combinations of tags and other
options.  Setting the bus to synchronous really killed performance.
Iozone, as I recall, gave a peak of about 2.5MB/sec.

Maybe there are some drive options I need to look at changing, possibly
to make synchronous SCSI work?  Anyone have suggestions for how to
make things likely go faster?  I'm hardly complaining, this is easily
the fastest desktop computer I've ever used... (*:

Anyhow, now I'm trying to upgrade to 2.0 from the walnut creek CDROM, but
not having any luck with my P90 (possible floppy drive problem, or device
interaction problem -- specific error comes trying to mount the root partition,
floppy driver reports it can't read cyl 0, track 0, sector 17 or around
there...).  I'll know more after this weekend.

Anyhow, many thanks to Stefan and Wolfgang for a job really well done!

Paul Anderson
pha@umich.edu