Linux_hpc_swstack - Sep 2008 - Feature Request: EDAC Utils

Hello, 

I would like to request that EDAC Utils working on today''s
Harpertown-based X6250s and tomorrow''s Nehalem-based blades be included
a future release of Sun(TM) HPC Software, Linux Edition:

        http://bluesmoke.sourceforge.net/
	http://sourceforge.net/projects/edac-utils

I need this software to be able to effectively deal with ECC memory
errors that will be fairly common at our scale, memory density, and
elevation. 

Thank you,

-Matthew

Memory errors on the x6250 are logged by the SP into the SEL, and tied 
into the blade''s fault reporting.
I think your request would need to clarify in which ways the existing 
tools are not adequate to identify
bad memory.

Kevin


Matthew Bohnsack wrote:
> Hello, 
>
> I would like to request that EDAC Utils working on today''s
> Harpertown-based X6250s and tomorrow''s Nehalem-based blades be
included
> a future release of Sun(TM) HPC Software, Linux Edition:
>
>         http://bluesmoke.sourceforge.net/
> 	http://sourceforge.net/projects/edac-utils
>
> I need this software to be able to effectively deal with ECC memory
> errors that will be fairly common at our scale, memory density, and
> elevation. 
>
> Thank you,
>
> -Matthew
>
> _______________________________________________
> Linux_hpc_swstack mailing list
> Linux_hpc_swstack at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/linux_hpc_swstack
>

Thanks for the pointer Kevin.

I will examine and evaluate the SP tools more closely over the next
weeks and get back to you on this.

-Matthew


On Fri, 2008-09-12 at 11:00 -0600, Kevin Van Maren
wrote:
> Memory errors on the x6250 are logged by the SP into the SEL, and
> tied 
> into the blade''s fault reporting.
> I think your request would need to clarify in which ways the existing 
> tools are not adequate to identify
> bad memory.
> 
> Kevin
>

Kevin and list,

We''re most interested in gaining this functionality for
Gainestown-based
Vayu blades.  We''re receiving 800+ of them very soon now.  It could be
true that memory error information is logged into the SEL via the SP,
but we haven''t received any concrete information on the capabilities of
the Vayu in this respect.  Could you provide this information to me?

Regardless, our experience doing this kind of thing on 1,000s of nodes
causes us to greatly prefer the EDAC approach for the following reasons:

     A. We have direct real-time access to the actual hardware events.
        With EDAC, we can get immediate visibility via syslog, when an
        ECC event happens, without having to deal with polling the SP''s
        SEL.  I.e., we want to know about ECC events the moment they
        happen.  Can the SP push all of these events to us via SNMP
        trap, syslog or another mechanism as they happen, or do we have
        to poll the SP? 
     B. We much prefer the direct access to memory fault information
        that EDAC provides over the filtered access that is usually
        provided by vendors in their SP/SEL implementations.  I.e., we
        want to control the error reporting thresholds and have access
        to all error data, instead of the data being filtered though
        opaque firmware containing thresholds that we can not change.
        Does the SP/SEL approach give me access to every event?  Can I
        modify all thresholds? 
     C. The use of EDAC is operationally consistent with all of our
        other platforms.

Does that make sense?

Thanks,

-Matthew

On Fri, 2008-09-12 at 11:09 -0600, Matthew Bohnsack wrote:
> Thanks for the pointer Kevin.
> 
> I will examine and evaluate the SP tools more closely over the next
> weeks and get back to you on this.
> 
> -Matthew
> 
> 
> On Fri, 2008-09-12 at 11:00 -0600, Kevin Van Maren wrote:
> > Memory errors on the x6250 are logged by the SP into the SEL, and
> > tied 
> > into the blade''s fault reporting.
> > I think your request would need to clarify in which ways the existing 
> > tools are not adequate to identify
> > bad memory.
> > 
> > Kevin
> > 
> 
> _______________________________________________
> Linux_hpc_swstack mailing list
> Linux_hpc_swstack at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/linux_hpc_swstack