[Swift-devel] Torus rank

[Emalayan Vairavanathan]

Kaz, Thank you very much for detailed clarification.

Regards
Emalayan



________________________________
 From: Kazutomo Yoshii <kazutomo at mcs.anl.gov>
To: Emalayan Vairavanathan <svemalayan at yahoo.com> 
Cc: "mosastore at googlegroups.com" <mosastore at googlegroups.com>; "swift-devel at ci.uchicago.edu" <swift-devel at ci.uchicago.edu> 
Sent: Wednesday, 7 March 2012 8:38 AM
Subject: Re: [Swift-devel] Torus rank
 

Thanks! Zhao

This is 99.99% guaranteed.  Let me explain why not 100%.

The control system loads personality (e.g. torus coordinate)
into memory on each node when they create a new partition via jtag.
No operating system runs at that point. The boot loader passes
the location of the personality to operating system (CNK or Linux).

NODE: personality information on node is a binary data of
the BGP personality struct; technically _BGP_Personality_t in
/bgsys/drivers/ppcfloor/arch/include/common/bgp_personality.h

Linux creates /proc/personality.sh by reading each member of
_BGP_Personality_t for convenience. CNK doesn't provide this.

If IBM change _BGP_Personality_t, personality.sh can be corrupted.
(probably won't happen but we can't guarantee this).

Linux also provides /proc/personality, which is a binary structure,
so you can can cast it to  _BGP_Personality_t, which is more reliable.

Due to other job pressure, I couldn't maintain all combinations.
However, I'm still updating zepto-vn-eval's compute node linux for
my kernel work.

As for the torus hardware rank,  I would definitely notice if that
gets corrupted because Linux kernel and BGP native MPI heavily
depends on personality.

Thanks,
Kaz

On 03/06/2012 03:22 PM, Emalayan Vairavanathan wrote:
> Great. Thank you Zhao.
> 
> ------------------------------------------------------------------------
> *From:* Zhao Zhang <zhaozhang at uchicago.edu>
> *To:* Emalayan Vairavanathan <svemalayan at yahoo.com>
> *Cc:* Kazutomo Yoshii <kazutomo at mcs.anl.gov>;
> "swift-devel at ci.uchicago.edu" <swift-devel at ci.uchicago.edu>; MosaStore
> <mosastore at googlegroups.com>
> *Sent:* Tuesday, 6 March 2012 12:41 PM
> *Subject:* Re: [Swift-devel] Torus rank
> 
> Hi, Emalayan
> 
> Emalayan Vairavanathan wrote:
>> Hi Kaz,
>>
>> I found the following about tours rank from
>> http://wiki.mcs.anl.gov/zeptoos/index.php/FAQ and want to double check
>> with you still this information valid.
>>
>>
>>      Torus rank:-
>>
>>
>>      A torus rank is a number identifying a compute node within a
>>      whole partition. In a way, it is much "nicer" than a pset rank
>>      since it is unique within a job and it also starts from 0.
>>
>> - A shell script can easily calculate it from other fields:
>> TORUS_RANK=`echo $BG_PSETORG | awk "{print \\$1 + \\$2 * $BG_XSIZE + \
>>            \\$3 * $BG_XSIZE * $BG_YSIZE}"
>>
>> It would be great if you can make sure that the tours rank is unique
>> within a partition and always starts from zero regardless of the
>> following.
>>
>>  - Partition size (fraction of pset / multiple of psets / both )
>>  - Platform (Surveyor / Intrepid)
>>  - Kernel profiles.
>>
> yes, this is true regardless of partition size, platform, and ZeptoOS
> kernel profiles(zeptoos, zepto-vn-eval, and the one you use for mosa).
> 
> best
> zhao
>>
>> (Note: I am asking this in order to figure out a easy way to deploy
>> MosaStore at BG/P at large scale. I did few tests with different
>> partition size in Surveyor and it seems the tours rank is unique and
>> always starts from zero.)
>>
>> Thank you
>> Emalayan
>>
>>
>> 
>> ------------------------------------------------------------------------
>>
>> _______________________________________________
>> Swift-devel mailing list
>> Swift-devel at ci.uchicago.edu <mailto:Swift-devel at ci.uchicago.edu>
>> https://lists.ci.uchicago.edu/cgi-bin/mailman/listinfo/swift-devel
>> 
> 
> -- 
> You received this message because you are subscribed to the Google
> Groups "MosaStore" group.
> To post to this group, send email to mosastore at googlegroups.com
> <mailto:mosastore at googlegroups.com>.
> To unsubscribe from this group, send email to
> mosastore+unsubscribe at googlegroups.com
> <mailto:unsubscribe at googlegroups.com>.
> For more options, visit this group at
> http://groups.google.com/group/mosastore?hl=en.
> 
> 
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/swift-devel/attachments/20120307/f2276b92/attachment.html>