From owner-freebsd-hackers  Thu Aug 31 17:51:33 1995
Return-Path: hackers-owner
Received: (from majordom@localhost)
          by freefall.FreeBSD.org (8.6.11/8.6.6) id RAA03521
          for hackers-outgoing; Thu, 31 Aug 1995 17:51:33 -0700
Received: from gndrsh.aac.dev.com (gndrsh.aac.dev.com [198.145.92.241])
          by freefall.FreeBSD.org (8.6.11/8.6.6) with ESMTP id RAA03515
          for <FreeBSD-hackers@freebsd.org>; Thu, 31 Aug 1995 17:51:31 -0700
Received: (from rgrimes@localhost) by gndrsh.aac.dev.com (8.6.11/8.6.9) id RAA13303; Thu, 31 Aug 1995 17:51:02 -0700
From: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com>
Message-Id: <199509010051.RAA13303@gndrsh.aac.dev.com>
Subject: Re: 4GB Drives
To: gibbs@freefall.FreeBSD.org (Justin T. Gibbs)
Date: Thu, 31 Aug 1995 17:51:02 -0700 (PDT)
Cc: FreeBSD-hackers@freebsd.org
In-Reply-To: <199509010034.RAA03131@freefall.FreeBSD.org> from "Justin T. Gibbs" at Aug 31, 95 05:34:16 pm
X-Mailer: ELM [version 2.4 PL24]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Content-Length: 2225      
Sender: hackers-owner@freebsd.org
Precedence: bulk

> 
> >> One thing to watch out for is that even without tagged queuing, you
> >> can get up to two requests per target sitting on the board.  You'll
> >
> >Those are the 2 I need to eliminate for my analysis.  Any quick dirty
> >hack I can do to turn them off?
> 
> Then set the number of commands that the scsi system is allowed to queue
> to 1 instead of 2.  Look on line 882 of aic7xxx.c:
> 
>         ahc->sc_link.opennings = 2;
> 
> Should be changed to 1.

THANKS!!  Finding that stuff buried in all the scsi code is painful.

> >> have to blink something when the command that gets taken off the queue
> >> is for a target that has an active command pending since it will re-queue
> >> the transaction in that case.  This will happen when the first transaction
> >> to the disk disconnects to do the seek.
> >
> >Test case has already eliminated all seeks.  Seeks do not occur as my total
> >transfer is smaller than 1 cylinder * stripe width.  I am trying to find
> >the proper parameters and emperical data to improve my theoritcal caclulations
> >and formulas.  I can't find the errors in these algorithms without some
> >correlation to real measured data to find the bad assumptions in them :-(.
> 
> Hmm.  Are you sure that the drive will not disconnect even for a short time
> after receiving its command?

Darn sure, I effectivly disabled disconnect/reconnect by tweaking the
buffer full/empty ratios and a few other parameters in mode page 2.

> >The data I need is being very difficult to obtain without spending major
> >dollars on a scsi bus analyzer and a large logic analyzer :-(.  I guess
> >I could go out to the labs some weekend, but it is a pain to check in this
> >much equipment.
> 
> I would think that a scope attached to the led line would be sufficient
> for most of the timing.

Yes, it should be, but if I have things like command queueing occuring I
can not correlate the edge created in the stripe layer with the edge
created by the LED, it may be skewed by 1 cycle.  Or I could go get a
$10k storage scope and use single events :-).


-- 
Rod Grimes                                      rgrimes@gndrsh.aac.dev.com
Accurate Automation Company                 Reliable computers for FreeBSD