NetBSD 3 SCSI broken

Hello,

After many months of testing things when I've had spare
time, I can now 
say definitively that Quadra class machines (605, 650, 800)
will 
eventually lock up under heavy SCSI usage. This does not
happen under 
NetBSD 1.6.x.

I've tried several different kinds of SCSI drives, with and
without 
command queueing, syncronous and asyncronous, the same
drives with 1.6 and 
3, and I have two long term test machines which have
significant uptime 
with 1.6.x and can be locked up by cvs'ing src and pkgsrc
at the same 
time.

It's hard to believe that there are so few people out there
that who 
either never thrash their disks or who don't observe the
same lockups. Or 
perhaps they think they're isolated or something. Not sure,
but it's 
definitely repeatable, and I've tried four different
machines with as many 
different hard drives now.

Ideas, anyone?

John

Hi,

Sadly I think that very few people use the mac68k systems
for long
times anymore, especially for servers.

In my case I use a portable, so 8 hrs or so is the max
uptime.  It's
too bad that it has broken though.

cheers

bruce

On Sun, Apr 16, 2006 at 01:51:08PM -0700, John Klos wrote:
> Hello,
> 
> After many months of testing things when I've had
spare time, I can now 
> say definitively that Quadra class machines (605, 650,
800) will 
> eventually lock up under heavy SCSI usage. This does
not happen under 
> NetBSD 1.6.x.
> 
> I've tried several different kinds of SCSI drives,
with and without 
> command queueing, syncronous and asyncronous, the same
drives with 1.6 and 
> 3, and I have two long term test machines which have
significant uptime 
> with 1.6.x and can be locked up by cvs'ing src and
pkgsrc at the same 
> time.
> 
> It's hard to believe that there are so few people out
there that who 
> either never thrash their disks or who don't observe
the same lockups. Or 
> perhaps they think they're isolated or something. Not
sure, but it's 
> definitely repeatable, and I've tried four different
machines with as many 
> different hard drives now.
> 
> Ideas, anyone?
> 
> John

-- 
edoneelsdf.lonestar.org
SDF Public Access UNIX System - http://sdf.lonestar.org

> Sadly I think that very few people use the mac68k
systems for long
> times anymore, especially for servers.

Well, I do, but mine are all 1.6.x since I haven't seen the
need to update.
This makes me even less likely to.

-- 
--------------------------------- personal: http://www.armory.com
/~spectre/ ---
  Cameron Kaiser * Floodgap Systems * www.floodgap.com *
ckaiserfloodgap.com
-- My Pink Floyd code: v1.2a s BO 1/0/pw tinG 0? 0 Relics 2
8 <6mar98> --------

On Mon, 17 Apr 2006 06:31:20 -0700 (PDT), Cameron Kaiser
<spectrefloodgap.com>, wrote:

>> Sadly I think that very few people use the mac68k
systems for long
>> times anymore, especially for servers.
>
>Well, I do, but mine are all 1.6.x since I haven't seen
the need to update.
>This makes me even less likely to.

   My LCIII has been up for 65 days now, running 2.1. Not
very long, but no
problems so far, either. 
--
Regards,
/Bjarne.

>> Sadly I think that very few people use the mac68k
systems for long
>> times anymore, especially for servers.
>
> Well, I do, but mine are all 1.6.x since I haven't
seen the need to update.
> This makes me even less likely to.

Exactly. Mine are excellent backup DNS and backup MX
servers, and are even 
doing things like helping to seed bittorrents. But no matter
how hard I 
try, I can't upgrade without ending up with machines which
lock up.

John

John Klos wrote:
>>> Sadly I think that very few people use the
mac68k systems for long
>>> times anymore, especially for servers.
>>
>> Well, I do, but mine are all 1.6.x since I haven't
seen the need to 
>> update.
>> This makes me even less likely to.
> 
> Exactly. Mine are excellent backup DNS and backup MX
servers, and are 
> even doing things like helping to seed bittorrents. But
no matter how 
> hard I try, I can't upgrade without ending up with
machines which lock up.

I've been building packages pretty regularly for several
months now, and 
haven't seen this locking up.  That machine uses 2.1.  But
I'm not sure 
that it reaches a level of disk usage that you'd consider
thrashing, 
either.  I've never really monitored it.

Tim

-- 
Tim & Alethea
christtrek.org

Hey,

On Sunday, April 16, 2006, at 10:51 PM, John Klos wrote:

> Hello,
>
> After many months of testing things when I've had
spare time, I can now 
> say definitively that Quadra class machines (605, 650,
800) will 
> eventually lock up under heavy SCSI usage. This does
not happen under 
> NetBSD 1.6.x.
>
> I've tried several different kinds of SCSI drives,
with and without 
> command queueing, syncronous and asyncronous, the same
drives with 1.6 
> and 3, and I have two long term test machines which
have significant 
> uptime with 1.6.x and can be locked up by cvs'ing src
and pkgsrc at the 
> same time.

I don't experience a lock-up, but something even more
nasty: corruption. 
I have a IIci which used to run 1.6 netbsd fine and  openbsd
3.3 
perfectly fine. Rock-stable. I then put in the hard-drive of
my q610 
which sadly died but which was rock-stable too, I
definitvely miss that 
beast!
Anyway, all my attempts to isntall netbsd 2.x or 3.0 (the
latest today) 
fail, in a bigger or smaller degree there is soemthing
unreliable in the 
scsi handlign. If I ftp and download a tar.gz it might
checksum wrongly 
on uncompression ! Once I was able to isntall a semi-usable
system and 
beside s noticing some strange RAM waste (but other reported
that and it 
may be solved) usage would suffer HD troubles and unreliable
files and 
even lock-ups (but I suspect that some binaries uncomrpesed
and 
installed wrongly, you can't be sure of anything).

unfortunate ly at the moment I don not have another spare HD
and also no 
t the time to parition, format and install both macos and
netbsd on it, 
especially since our installer is still broken.

-R

At 13:51 Uhr -0700 16.4.2006, John Klos wrote:
>After many months of testing things when I've had spare
time, I can now
>say definitively that Quadra class machines (605, 650,
800) will
>eventually lock up under heavy SCSI usage. This does not
happen under
>NetBSD 1.6.x.

As a late follow-up:

If you can reproduce this pattern, please collect evidence
and submit a PR,
so that mac68k goes on record with the problem.

Thanks,
	hauke

--
"It's never straight up and down"     (DEVO)