Re: [Freeipmi-devel] sun20 workaround doesn't

[Al Chu]

Hey Dave,

> OK, appended.

Hmmm.  It seems the workaround just isn't working.  It's possible I have
introduced a bug into the workaround.  Let me see if I can figure it
out.

> I didn't mean to be critical, of course.  However, I find it's a real
> practical problem with our heterogeneous HPC systems, and I guess I'm
> not alone.  I notice there's a TODO item addressing workarounds per
> host.  Is it clear what's involved if I give it a go?

Hehe.  It's funny you mention this.  I had a request/comment on this
topic just last week.  First, please take a look at my original post on
this for what I was thinking of supporting:

http://www.mail-archive.com/address@hidden/msg00710.html

The work involved is somewhat deep (that I've naturally been putting
off, but it is now higher priority b/c it came as a request from someone
internal to my organization).  As far as I can tell, it would take:

A) re-work a lot of the config file parsing code to be able to take an
additional argument.  Based on other people's requests, this would be
for all the config file options.  It could be made easier if instead of
modifying all config file options to take the additional argument, we
made new config file options, like:

# default for all systems
workaround intel20
# default for some spsecific systems
workaround-hosts node[0-11] sun20

I haven't decided which I will do at this moment.

B) modifying the tools to override it's defaults + the config file
defaults w/ the host specific defaults.  This will require a fair amount
of re-architecture b/c host-specific defaults need to be stored/passed
around/parsed "deeper" into the hostrange code.

Al

On Tue, 2009-06-09 at 10:43 +0100, Dave Love wrote:
> Al Chu <address@hidden> writes:
> 
> > Hey Dave,
> >
> > It's entirely possible there is another issue for your systems I never
> > encountered before or a bug in the workaround code.  If you could send
> > the --debug output, that'd be great.
> 
> OK, appended.
> 
> > Ipmitool also elects to "hide"
> > workarounds in its tools more often, instead of making the user put them
> > on the command line.  Pros and cons of both approaches, I don't think
> > either of us is right or wrong.
> 
> I didn't mean to be critical, of course.  However, I find it's a real
> practical problem with our heterogeneous HPC systems, and I guess I'm
> not alone.  I notice there's a TODO item addressing workarounds per
> host.  Is it clear what's involved if I give it a go?
> 
-- 
Albert Chu
address@hidden
Computer Scientist
High Performance Systems Division
Lawrence Livermore National Laboratory