poster FDSI SPADES.pptx - Agence Nationale de la Recherche

FONDEMENT DES SYSTÈMES
INFORMATIQUES
SPADES:
Servicing Petascale Architectures
and Distributed System
SEGI 2008
Overview!
Today's emergence of Petascale
architecture and the evolution of both
research and production grids
increase a lot the number of potential
r e s o u r c e s . H o w e v e r, e x i s t i n g
infrastructures and access rules do
not allow to fully take advantage of
these resources. One key idea of the
SPADES project is to propose a nonintrusive but highly dynamic
environment able to take advantage of
available resources without disturbing
their native use. "
Goals:!
•  Support platforms at a "
very large scale (Petas-"
cale environments"
Means: !
•  Distributed Schedulers "
based on the Desktop "
Computing paradigm."
Co-scheduling0
The$scheduling$
en--es$($$$$$$$$$$$$$$)$$
Scheduler0
Scheduler0
have$to$cooperate$$
with$both$Service$
Discovery$Agents$($$$$$$$$$)$$
SDA0
and$Batch$Schedulers$($$$$$$$$)$
BS0
ObjecGves:0
• $Determining$the$
BS0
Best$mapping$of$the$
jobs$on$resources.$
• $Designing$reac-ve$$
Scheduling$heuris-cs.$$
Approaches:0
• $Adap-ng$strategies$coming$from$
Desktop$Compu-ng$$
• $Replica-on$strategies$$
• $Coopera-on$between$several$batch$
schedulers$$
• $OnEdemand$job$migra-on$
COORDINATEURS:
PARTENAIRES:
[email protected]
Tel:004.72.72.80.040
h9p//graal.ens-lyon.fr/SPADES0
P2P Service Discovery in a large
scale and distributed context."
Constraints:!
•  Platforms heterogeneity"
•  System dynamicity"
Objectives:!
•  Reliability of infrastructure"
•  Efficiency of" SDA0
Fault-tolerance mechanisms:!
•  Self-stabilisation"
•  Replication. "
Petascale Runtimes:
Runtimes hitherto proposed do
not fit any more with the 100.000
resources
Problems:
•  Deployment and launching
step duration
•  Scaling data structures
•  Failure detection broadcasting
•  Failure recovery
Issues:
•  Unifying resources
•  Efficient communication
Means:
•  P2P approach
•  Overlay networks
BS0
Expected results:
-  n u m e r o u s a n d
frequent failures
Scheduler0
SDAs0
BS0
BS0
Batch Scheduler Interactions!
To allow reactive Petascale architectures, fine interactions between
Batch Schedulers ( BS0 ) and upper
layers is a key issue."
Approachs: !
•  Using standards API (OGSA-BES,
DRMAA) hiding heterogeneity."
•  Customizing a BS and executing it in
parallel with other BS."
Applications
Target applications:
•  Cosmic ray simulator
•  Coupled climate model.
Test platform:
•  IBM Blue Gene
•  CRAY XT of PRACE
•  DEISA grids
Benefits:
•  Increase model resolution
•  Reduce model time step
•  Make replications
Expected results: Observe low scale
Phenomena as hurricanes.