Abstract:
The MPI Standard deliberately does not make any performance guarantees,
but users expect MPI implementations to deliver good performance that is
in some sense portable across platforms and MPI implementations. In this
talk we formulate and discuss performance guidelines and conditions that
good MPI implementations can be expected to fulfill by relating aspects
of the MPI standard to each other. We claim that such self-consistent
performance guidelines help toward achieving performance portability.
The formulation could be used by benchmarks and tools, such as SKaMPI
and Perfbase, to automatically verify whether a given MPI implementation
fulfills basic performance guidelines. We present examples where some of
these requirements are not satisfied, demonstrating that there remains
room for improvement in MPI implementations.
Zeit: Donnerstag, 30.10.2008, 16h ct.
Ort: Seminarraum C316
Universität Wien
Institut für Scientific Computing
Nordbergstraße 15/C/3
1090 Wien
<http://www.par.univie.ac.at/>
Da das hier eine Diskussionsgruppe ist gebe ich einmal Feedback:
Sabri Pllana <pll...@par.univie.ac.at> writes:
> Vortragstitel: Self-consistent MPI Performance Guidelines
> Dr. Jesper Larsson Träff
> NEC Laboratories Europe
> St. Augustin, Germany
>
> Abstract:
> The MPI Standard deliberately does not make any performance
> guarantees, but users expect MPI implementations to deliver good
> performance that is in some sense portable across platforms and MPI
> implementations. In this talk we formulate and discuss performance
> guidelines and conditions that good MPI implementations can be
> expected to fulfill by relating aspects of the MPI standard to each
> other. We claim that such self-consistent performance guidelines help
> toward achieving performance portability. The formulation could be
> used by benchmarks and tools, such as SKaMPI and Perfbase, to
> automatically verify whether a given MPI implementation fulfills basic
> performance guidelines. We present examples where some of these
> requirements are not satisfied, demonstrating that there remains room
> for improvement in MPI implementations.
Ich habe die Beschreibung bis hierher gelesen und weiß immer noch
nicht, worum es eigentlich geht; eigentlich geht nicht einmal klar
heraus, um welches Themengebiet es sich handelt (vermutlich irgendwas
mit Software, da von Portabilität die Rede ist). Nur mal so als
Anregung, da mit dem Posting in at.tuwien.student offenbar ein breites
Publikum angesprochen werden soll...
LG,
Heinzi
Hier findet man eine Übersicht über
Message Passing Interface (MPI)
http://de.wikipedia.org/wiki/Message_Passing_Interface
LG,
Sabri Pllana
--
<http://www.par.univie.ac.at/~pllana/>
Danke!
LG,
Heinzi
FREE USPS PRIORITY MAIL! BONUS SOFTWARE! CASHBACK!
Click to go
HELLO ALL AND WELCOME TO OUR BUY IT NOW FOR A BRAND NEW, 100% AUTHENTIC AND
SEALED APPLE IPOD TOUCH! ,
8GB MODEL!! This is the latest and greatest from Apple and the 2ND
Generation of this wonderful iPod!
As well, this is eligible for live.com's cashback. You can get this iPod for
about $186!
Please email us if you need assistance on getting the cashback rebate.
Click to go
"Sabri Pllana" <pll...@par.univie.ac.at> ha scritto nel messaggio
news:4905887b$0$11868$3b21...@usenet.univie.ac.at...
>
> Vortragstitel: Self-consistent MPI Performance Guidelines
> Dr. Jesper Larsson Tr�ff
> NEC Laboratories Europe
> St. Augustin, Germany
>
> Abstract:
> The MPI Standard deliberately does not make any performance guarantees,
> but users expect MPI implementations to deliver good performance that is
> in some sense portable across platforms and MPI implementations. In this
> talk we formulate and discuss performance guidelines and conditions that
> good MPI implementations can be expected to fulfill by relating aspects of
> the MPI standard to each other. We claim that such self-consistent
> performance guidelines help toward achieving performance portability. The
> formulation could be used by benchmarks and tools, such as SKaMPI and
> Perfbase, to automatically verify whether a given MPI implementation
> fulfills basic performance guidelines. We present examples where some of
> these requirements are not satisfied, demonstrating that there remains
> room for improvement in MPI implementations.
>
> Zeit: Donnerstag, 30.10.2008, 16h ct.
> Ort: Seminarraum C316
> Universit�t Wien
> Institut f�r Scientific Computing
> Nordbergstra�e 15/C/3
> 1090 Wien
> <http://www.par.univie.ac.at/>