Subject: Re: Sparc advice.
To: None <port-sparc@NetBSD.org>
From: Steve Rikli <sr@genyosha.net>
List: port-sparc
Date: 05/05/2005 17:41:08
On Thu, May 05, 2005 at 06:32:10PM -0400, Christopher Sean Hilton wrote:
> 
> I have an SS20 with a 180 MHz Ross Hypersparc CPU. I would like to use
> this machine as a webserver which I will remotely locate in a friends
> closet. The machine has two 1G internal drives and a 4G Seagate Hawk in
> an external case. I have a Sun Fast Wide SCSI Interface and Happy Meal
> Ethernet card in the box. It's got 512MB of RAM and 512MB of swap. I'd
> like to offer my compliments to the NetBSD team. I'm very positively
> impressed by NetBSD on this hardware.
> 
> I have a couple of questions.

I can maybe help answer a little ...

> Firstly I'm  having problems with the long
> term stability of the machine. I'm getting:
> 
>    sbrk grow <number>: error 12
> 
> messages after the machine has run for a while. These problems start
> when I compile something from source in /usr/pkgsrc or lately when I run
> emacs to edit a web page. I've seen some posts on this problem before
> and I'm just wondering if this is something that I will have to upgrade
> to -current to fix.

If the posts you've seen are in the same vicinity of the "kmem_map"
panics which some of us w/larger RAM in our SPARCs were seeing, you
might get decent results by bumping up NKMEMPAGES in your kernel.

I have a SS20 w/dual 180 Ross & 512MB running 2.0, and the system
stabilized pretty much completely with this added to kernel conf:

  options         NKMEMPAGES=5120

I too would love to know if that's something which might not be
required in future releases.  I haven't un-done that setting for some
time, since the box has been very well behaved after adding it, but
it looks to my untrained eye like the default is still 1536 (6MB).

> Second: make search key=foo in /usr/pkgsrc doesn't work because the box
> doesn't have the pkgsrc index built. How long can I expect a make index
> to run and is there anyway to performance tune/optimize this process?

Haven't tried that yet -- I guess we'll see.  :-)

> Third: I've been thinking about getting another 180 MHz CPU but I'm
> concerned about heat. When I've gone into the case of this box after
> it's been running I've noticed that the CPU heatsink is warm to the
> touch, not hot, just warm. Would running two of these CPU's lower the
> reliablity of this machine? Put a different way, is the Case up to
> cooling that combination?

IME, not really.  It'll depend somewhat on the environment of your
friend's closet -- if it's a pretty well air-conditioned space with
good circulation, you could give it a shot.  Definitely make sure
that the case fans are clean and otherwise unobstructed by dust and
other old machineroom "gunk" -- doing that will help some.  I cooled
off a SS5 a bit just by giving the case fans a good spritzing.  :)

My SS20/dual-180 lives in a room-temperature rack (along with a
couple other SPARCs) where the primary air circulation in the room
is a ceiling fan plus a floor fan during the summer.  Thus, I added
extra little fans into the SS20, 1 where the CD player would be,
blowing on the disk drives, and another towards the front of the
chassis, opposite the case fans, to keep air flowing across the CPUs.

With those, the insides seem to stay reasonably warm, but not hot.
Ross get pretty toasty anyway, but now you could touch them without
burning your hand, and the drives aren't melting either.  Yet.  ;-)

cheers,
sr.
-- 
|| Steve Rikli              ||| You can use all the quantitative data you ||
|| Systems Administrator    ||| can get, but you still have to distrust   ||
||                          ||| it and use your own intelligence and      ||
|| sr@genyosha.net          ||| judgment.              - Alvin Toffler    ||