Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 14 Apr 1995 19:48:39 -0400 (EDT)
From:      Peter Dufault <dufault@hda.com>
To:        bde@zeta.org.au (Bruce Evans)
Cc:        bakul@netcom.com, rgrimes@gndrsh.aac.dev.com, freebsd-scsi@FreeBSD.org
Subject:   SCSI command overhead
Message-ID:  <199504142348.TAA03880@hda.com>
In-Reply-To: <199504142216.IAA23515@godzilla.zeta.org.au> from "Bruce Evans" at Apr 15, 95 08:16:59 am

next in thread | previous in thread | raw e-mail | index | archive | help
Bruce Evans writes:
> 
> >> On a somewhat related topic, does anyone know command
> >> processing overhead of modern SCSI devices?  It used to be
> >> in the 50 ~ 100 us range a few years ago.
> 
> >Not sure what it is, but most of my Quantum disk drives are using
> >a 25Mhz 68020 and a Qlogic scsi bus controller chip, should be much
> >much faster than the old stuff that was Z80 based :-)
> 
> It seems to be much larger than 50 ~ 100 us :-(.  According to the
> following test program, it is about 3500 us for a Buslogic BT445C
> controller / Toshiba MK537FB drive and about 2800 us for a BT445C
> controller / Quantum XP34301 drive.

I'd like to work out a careful breakdown of where the overhead is.  Bruce
is in the right ballpark.  Using libscsi I just put together a loop of
1000 "test unit readies" and sent it to the three disks I have here
when the system was idle (verified by seeing that the disks were always
ready).  Then I stepped all over the CDB so that the driver would reject
it to get a feel for the overhead of the SCSI code down to the point it
was about to submit the CDB to the host adapter.  And it is awful
(This is with a 1542C):

hda# ./tur /dev/rsd0c
CDC     94171-9: 2535 microseconds.
                 92 microseconds.
hda# ./tur /dev/rsd1c
FUJITSU M2266S-5: 2324 microseconds.
                  85 microseconds.
hda# ./tur /dev/rsd2c
FUJITSU M2654S-5: 2280 microseconds.
                  87 microseconds.

So although we're seeing almost a 300 microsecond improvement between
the oldest and the newest device, but the overhead is still more than 2
milliseconds per transaction.

-- 
Peter Dufault               Real Time Machine Control and Simulation
HD Associates, Inc.         Voice: 508 433 6936
dufault@hda.com             Fax:   508 433 5267



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199504142348.TAA03880>