From bonniema@APG.CA  Tue Aug 15 12:09:12 1995
Received: from apg.ca (pub.apg.ca [204.191.66.254])
          by freefall.FreeBSD.org (8.6.11/8.6.6) with ESMTP id MAA20684
          for <FreeBSD-gnats-submit@freebsd.org>; Tue, 15 Aug 1995 12:09:05 -0700
Received: (from bonniema@localhost) by apg.ca (8.6.11/8.6.9) id PAA02120 for FreeBSD-gnats-submit@freebsd.org; Tue, 15 Aug 1995 15:12:41 -0400
Message-Id: <199508151912.PAA02120@apg.ca>
Date: Tue, 15 Aug 1995 15:12:41 -0400
From: Marc Bonnier  <bonniema@APG.CA>
To: FreeBSD-gnats-submit@freebsd.org
Subject:

>Number:         691
>Category:       kern
>Synopsis:       Performance 10 times slower 2.0.5R & Adaptec AHA-2842.
>Confidential:   no
>Severity:       critical
>Priority:       high
>Responsible:    gibbs
>State:          closed
>Quarter:
>Keywords:
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Tue Aug 15 12:10:01 PDT 1995
>Closed-Date:    Wed Feb 7 06:26:06 PST 1996
>Last-Modified:  Wed Feb  7 06:27:19 PST 1996
>Originator:     Marc Bonnier
>Release:        FreeBSD 2.1.0-950726-SNAP
>Organization:
APG Technologies Inc.
>Environment:
>Description:
    Upgrading from
        "FreeBSD 2.1.0-Development #0: Sun Dec  4 17:30:34 GMT 1994"
    to
        "FreeBSD 2.1.0-950726-SNAP #0: Fri Aug 11 04:25:23  1995"

    ** The general performance drop by 10 times :( !! ** Disk IO and
    paging appear to be elapsed fast burst transfert.

    [ A qualitative gut feeling about the sound of my disk ): ] 


    Benchmark:
      Rebuilding the kernel (make) on 2.1.0-Dev:
        774.3u 77.8s 14:46.76 96.1% 1023+1274k 1888+5296io 184pf+0w

      Rebuilding the kernel (make) on 2.1.0-950726-SNAP:
        7239.2u 252.9s 2:07:55.40 97.6% 1023+1309k 1406+3430io 66pf+0w

      Rebuilding the kernel (make) on 2.1.0-950726-SNAP 
      *WITH* old aic driver:
        735.8u 63.5s 15:05.51 88.2% 1063+1322k 1640+3956io 39pf+0w


>How-To-Repeat:
    Repeatable 99% of time.

>Fix:
    In all the changes made in 2.0.5R, the Adaptec new aic driver is
    the culprit.
   
  FreeBSD 2.1.0-Development-1994 (old aic driver & sequencer)
    ./sys/gnu/misc/aic7770
    ./sys/gnu/misc/aic7770/aic7770.1
    ./sys/gnu/misc/aic7770/aic7770.c
    ./sys/gnu/misc/aic7770/aic7770.seq
    ./sys/gnu/misc/aic7770/aic7770_seq.h
    ./sys/i386/isa/aic6360.c
    ./sys/i386/isa/aic7770.c
  
  FreeBSD 2.1.0-950726-SNAP (new driver & sequencer)
    ./sys/dev/aic7xxx
    ./sys/dev/aic7xxx/aic7xxx.seq
    ./sys/dev/aic7xxx/aic7xxx_asm.1
    ./sys/dev/aic7xxx/aic7xxx_asm.c
    ./sys/i386/isa/aic6360.c
    ./sys/i386/isa/aic7770.c
    ./sys/i386/scsi/aic7xxx.c
    ./sys/i386/scsi/aic7xxx.h
    ./sys/pci/aic7870.c

Fix #1 As A WORKAROUND (use the old driver & old sequencer)

  1) With the "2.1.0-950726-SNAP" source tree.

  2) Use the "2.1.0-Development-1994" sub-tress for:
    ./sys/gnu/misc/aic7770/aic7770.c
    ./sys/gnu/misc/aic7770/aic7770.seq
    ./sys/gnu/misc/aic7770/aic7770_seq.h
    ./sys/i386/isa/aic7770.c

  3) Build the kernel
    config DELL
    cd ../../compile
    vi Makefile
      delete the "aic7xxx.o" from OBJ variable.
    make depend
    make
    reboot the new kernel

    Benchmark:
      Rebuilding the kernel (make) on 2.1.0-950726-SNAP with OLD AIC driver:
        1104.9u 96.6s 20:53.94 95.8% 118+684k 1401+79io   97pf+0w

-------------------------
Thank to all member of the team and all contributors to make it
happend!

Marc Bonnier
APG Technologies
Marc.Bonnier@APG.CA
>Release-Note:
>Audit-Trail:
State-Changed-From-To: open->analyzed 
State-Changed-By: gibbs 
State-Changed-When: Sat Jan 6 10:42:02 PST 1996 
State-Changed-Why:  
This may have to do with the setting of the BUSSPEED register 
on the 284x.  I'll reinstall my 284x later today and see if I can 
reproduce the problem with the -current driver.  It would be 
interesting to know if you can reproduce this with -stable or 
-current as well. 
Responsible-Changed-From-To: gnats-admin->gibbs 
Responsible-Changed-By: pst 
Responsible-Changed-When: Tue Feb 6 22:06:40 PST 1996 
Responsible-Changed-Why:  
State-Changed-From-To: analyzed->closed 
State-Changed-By: gibbs 
State-Changed-When: Wed Feb 7 06:26:06 PST 1996 
State-Changed-Why:  
Not reproducable.  My 2842 is actually slightly faster than the 
2742 sitting right next to it in my system. 
>Unformatted:
 >Environement:
     DELL Optiplex 466/LV
     486DX266 CPU 256KB Cache
     16 MB 
     Adaptec AHA-2842A (VLB RISC AIC7 )
     TSENG ET4000 W32i SVGA graphic acc.
     Seagate ST12550N
 
