From owner-freebsd-current  Sun Feb 18 18:59:22 1996
Return-Path: owner-current
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id SAA22091
          for current-outgoing; Sun, 18 Feb 1996 18:59:22 -0800 (PST)
Received: from luke.pmr.com (luke.pmr.com [206.224.65.132])
          by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id SAA22081
          Sun, 18 Feb 1996 18:59:14 -0800 (PST)
Received: (from bob@localhost) by luke.pmr.com (8.6.12/8.6.9) id UAA02614; Sun, 18 Feb 1996 20:59:11 -0600
From: Bob Willcox <bob@luke.pmr.com>
Message-Id: <199602190259.UAA02614@luke.pmr.com>
Subject: Re: Exabyte 8mm tape drive performance in -current?
To: gibbs@freefall.freebsd.org (Justin T. Gibbs)
Date: Sun, 18 Feb 1996 20:59:11 -0600 (CST)
Cc: freebsd-current@freefall.freebsd.org
In-Reply-To: <199602190132.RAA17788@freefall.freebsd.org> from "Justin T. Gibbs" at Feb 18, 96 05:32:57 pm
X-Mailer: ELM [version 2.4 PL24]
Content-Type: text
Sender: owner-current@FreeBSD.ORG
Precedence: bulk

Justin T. Gibbs wrote:
> 
> >I have observed that the (read and write) performance of my 8mm
> >Exabyte tape drives on my -current system runs roughly half of what
> >it is on my 2.1-stable systems (100kb/sec vs. 200kb/sec).  This is
> >with both the NCR 810 and Adaptec 2940 adapters and using programs
> >such as dump, tar, dd, team.  The systems that I have compared have
> >roughly the same hardware (both are 100MHz Pentiums).  Performance
> >on my Wangtek QIC-525 tape drive is about the same.  Can anybody
> >offer up an explaination of why this is and what might be done to
> >fix it?
> 
> I'd like to see if I can reproduce this on my system here.  I have
> both a -stable and -current boot configuration so I can test on
> both.  I'm using an Archive Python though, but if this is a generic
> SCSI problem, it should still show up.  Do you have a particular 
> benchmark script you'd like me to use?

For the sake of simplicity, I have just been using dd.  I dd a
large (>40mb) file to the tape using a bs= specification of 32k or
so.  Dump, and tar show similar performance differences.

My most recent run on my -current system looks like this:

bob@han-p1 /usr/src> dd bs=32k if=cscope.out of=/dev/rst2 conv=sync
1369+1 records in
1370+0 records out
44892160 bytes transferred in 405 secs (110844 bytes/sec)

and for this drive:

bob@han-p1 /home/bob> mt -f /dev/rst2 status     
Present Mode:   Density = 0x00         Blocksize = 512 bytes
---------available modes---------
Mode 0:         Density = 0x00         Blocksize = 512 bytes
Mode 1:         Density = X3.136-1986  Blocksize = 512 bytes
Mode 2:         Density = X3.39-1986   Blocksize variable
Mode 3:         Density = X3.54-1986   Blocksize variable

is what mt status looks like.  Note that I am unable to change the
blocksize to 0 (or anything else) with the mt blocksize command.


For the equivalent operation on one of my 2.1-stable systems I get:

bob@luke-pd /usr/src> dd bs=32k if=cscope.out of=/dev/rst1 conv=sync 
1334+1 records in
1335+0 records out
43745280 bytes transferred in 213 secs (205376 bytes/sec)

with an mt status of:

bob@luke-pd /usr/src> mt -f /dev/rst1 status
Present Mode:   Density = 0x00         Blocksize variable
---------available modes---------
Mode 0:         Density = 0x00         Blocksize variable
Mode 1:         Density = 0x00         Blocksize variable
Mode 2:         Density = 0x00         Blocksize variable
Mode 3:         Density = 0x00         Blocksize variable


Thanks,
-- 
Bob Willcox
bob@luke.pmr.com
Austin, TX