Date: Sat, 6 Nov 1999 10:24:11 +0100 (MET) From: Gerard Roudier <groudier@club-internet.fr> To: Ollivier Robert <roberto@keltia.freenix.fr> Cc: scsi@FreeBSD.ORG Subject: Re: SYM driver available for -stable Message-ID: <Pine.LNX.3.95.991106094852.360A-100000@localhost> In-Reply-To: <19991106010103.A2215@keltia.freenix.fr>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 6 Nov 1999, Ollivier Robert wrote:
> Gerard Roudier disait :
a os=E9 =E9crire : :-)
> > sym_conf.h file. You may edit this file, define SYMSETUP_SCSI_LED to 1 =
and
> > let us know if the 810a now is just fine for your next Christmas tree. =
:)=20
>=20
> Doesn't seem to make any difference. The light (for the first controller =
that
> is the 875) is still on permanently.
Will check the code. The sym does additionnal checking against some
initial IO registers and may be this initial IO register value (GPCNTL) is
not tested correctly or was not set as expected by BIOS.=20
> > 1 (no tags), 4, 8, 16.
>=20
> Here are the results:
>=20
> NCR / SYM with 1, 4, 8 and 16 tags
>=20
> -------Sequential Output-------- ---Sequential Input-- --Ran=
dom--
> -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --See=
ks---
> Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec=
%CPU
> ncr-1 200 3098 40.9 2998 12..7 1803 8.1 5791 48.4 5671 14.8 126.7=
4.1
> sym-1 200 3047 38.7 2983 11.6 1840 7.4 5715 46.2 5839 13.5 125.3=
3.9
>=20
> ncr-4 200 4589 60.2 4658 18.7 1788 8.3 5745 48.5 5794 15.7 130.7=
4.5 =20
> sym-4 200 4587 58.4 4660 17.0 1896 8.0 4911 39.8 5613 13.4 132.0=
4.3
>=20
> ncr-8 200 4670 61.3 4887 19.5 1900 8.9 5736 48.0 5840 15.6 130.0=
4.5
> sym-8 200 4811 61.2 4849 17.7 1947 8.3 5729 46.4 5596 13.5 129.2=
4.1
>=20
> ncr-16 200 4687 60.9 4916 19.9 1867 8.8 5748 48.0 5839 15.6 131.3=
4.5
> sym-16 200 4802 60.8 4697 17.1 1960 8.1 5192 41.9 5785 14.0 126.0=
4.2=20
The only differences I expected are CPU load and may-be rewrites.
- CPU because the sym driver uses O(1) everywhere. But I am surprised to=20
actually see a noticeable difference with a simple Bonnie benchmark.
- Rewrite because:
* lower latency may help device caching in this sitiation.
* and the sym driver allocates tags in a circular manner (just a differen=
ce=20
that seems to be a more soft policy for firmwares, especially broken
ones)
Other results should not be significantly different given the low
throughput and low Transactions/Second that should allow the device to
overlap latencies the same way for the both drivers.=20
The 5745/4911 and also the 5748/5192 read char results are extremally
bogus IMO, since they not only correspond to -low throughtput/large IOs-
cases but the kernel also perform prefetching (asynchronous readahead). A
driver bug is obviously possible, but I would be very surprised of such a
cause.=20
May-be, some Vinum array using something like 3..6 ultra-fast disks should
be used for these kind of comparisons.
Thanks a lot for your testing. I will try to understand the LED issue and
let you know when I have some diffs.
BTW, I donnot see any bad effect of using reasonable numbers tags with
your hard disk under Bonnie.=20
Regards,
G=E9rard.
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-scsi" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.LNX.3.95.991106094852.360A-100000>