[auton-users] [Fwd: Not entirely happy] (fwd)

[Paul Komarek]

Hi Chris,

The problem we're having with IDL is that the GNU/Linux Alpha version
keeps triggering unaligned exceptions.  It's not completely clear whether
these are the fault of IDL or something in Linux (the kernel).  That said,
IDL is currently causing (directly or indirectoy) many hundreds of these
exceptions right now, which explains your slow-down.

I'm not sure what to do or say.  Here are some options:

1) Changing Linux kernels has helped in the past (making it look like the
kerenls are guilty).  We're already current with the kernels, but could
try an older kernel (while hopefully avoiding the weekly-scsi-error-fest
we had with many of the older kernels).

2) We could reinstall Tru64 on Limey.  However, nobody seemed to like
Tru64 -- limey/Tru64 was only used for occasional IDL and scp tasks over
the last several months and never registered any significant or continued
load (except for some of my doing).

3) We could complain to the IDL folks.  They have a new version of IDL out
for every platform they support (it seems), except for GNU/Linux on Alpha.

I'd appreciate suggestions.  For the record, #2 is my least favorite
because

a) I'd like to standardize on GNU/Linux for the Alphas and hence avoid the
hours of extra work I've put in adapting admin stuff to Tru64, not to
mention changing my personal stuff from time to time as well as the AUTON
lab's build scripts,

b) It seems a big waste to use limey only for IDL and scp, as it has been
used for the last several months with Tru64 (even when the GNU/Linux
Alphas have been slammed).  We bought for over $40,000 and currently sells
for over $10,000.  Limey has a nice big 64 bit address space, 4GB of ram,
and a 4MB cache.  Sustaining an average load of 0.00 is a bit silly.

-Paul

On Tue, 17 Dec 2002, Paul Komarek wrote:

>
>
> ---------- Forwarded message ----------
> Date: Tue, 17 Dec 2002 16:20:16 -0500
> From: Chris Miller
> To: Paul Komarek <komarek at andrew.cmu.edu>
> Subject: [Fwd: Not entirely happy]
>
>   Paul, I was unable to send to user at autonlab.org from my machine
> (I guess I need the whole, unaliased group)..
>
> Could you forward this to the mailing list?
>
>
> -------- Original Message --------
> Subject: Not entirely happy
> Date: Tue, 17 Dec 2002 16:17:34 -0500
> From: Chris Miller <chrism at cmu.edu>
> Reply-To: chrism at cmu.edu
> To: user at autonlab.org
>
>
>
> Dear Paul  and others,
>   Something definitely changed on the auton linux-alphas.
> My code seems to run faster than it use to. But note that it
> never runs fast (i.e. it is typically as fast as on my 1Ghz Pentium).
>
> However, I started a code earlier today and noticed something
> had changed within the last couple of hours: I went from running
> pretty fast, to a dead crawl. I have suspended the code (while
> still in IDL) and when I type: print, sin(whatever), it takes a full second
> to calculate.  That same command runs instantaneously on Crazy.
>
> It *wasn't* having the problem a couple of hours ago (i.e. the
> code got about half-way done before this started happening).
>
> Any ideas? Anything in the kernel logs look funny?
>
> Chris
>
>
>
>
>
>
>