18th Workshop on Job Scheduling Strategies for Parallel Processing (JSSPP)
In Conjunction with
IPDPS 2014
Phoenix, Arizona, USA
23 May 2014
Workshop organizers
Walfredo Cirne, Google
Narayan Desai, Argonne National Laboratory
Program Committee
Henri Casanova, University of Hawaii at Manoa
Julita Corbalan, Technical University of Catalonia
Dick Epema, Delft University of Technology
Gilles Fedak, INRIA
Dror Feitelson, The Hebrew University
Liana Fong, IBM T. J. Watson Research Center
Eitan Frachtenberg, Facebook
Ali Ghodsi, UC Berkeley
Alfredo Goldman, University of Sao Paulo
Allan Gottlieb, New York University
Alexandru Iosup, Delft University of Technology
Morris Jette, SchedMD LLC
Rajkumar Kettimuthu, Argonne National Laboratory
Dalibor Klusáček, Masaryk University
Zhiling Lan, Illinois Institute of Technology
Bill Nitzberg, Altair Engineering
Larry Rudolph, MIT
Uwe Schwiegelshohn, Technical University Dortmund
Mark Squillante, IBM T. J. Watson Research Center
Murray Stokely, Google
Wei Tang, Argonne National Laboratory
Dan Tsafrir, Technion - Israel Institute of Technology
Ramin Yahyapour, GWDG - University of Göttingen
Keynote:
Requirements and Challenges of Workload Management in Cognitive Computing Liana Fong, IBM TJ Watson Research Center
It is difficult to define key characteristics of cognitive computing
systems while these systems are still being developed, their potential
capabilities are being expanded, and possible applicability is being
explored. However for now, we would attempt to define that cognitive
systems have the operation capabilities of four “I”s (eyes!):
Ingestion and Interpretation of data of high volume, velocity, variety
and veracity (the four Vs); Interaction with humans using means
(e.g. natural languages) beyond the classical text input/output, and
Insights producing to aim users’ decision making or/and actions.
Using some exemplary cognitive applications, we would show the
operations of the four “I”s are both data and/or compute intensive,
but the execution models and QoS are different for the “Is” and thus
impose different requirements in workload managements. There exists
also demand to support multi-user and multi-tenancy of cognitive
services in the cloud. In this talk, we examine the similarity and
difference in workload management for each “I”, and between cognitive
system & non-cognitive applications in clustered environment. Then,
we highlight the challenges and necessary novel technologies in the
workload management for cognitive systems.
Priority Operators for Fairshare Scheduling Gonzalo P. Rodrigo (Umeå University, Sweden); Per-Olov Östberg (Umeå
University, Sweden); Erik Elmroth (Umeå University, Sweden)
Dynamically Scheduling a
Component-Based Framework in Clusters Aleksandra Kuzmanovska (Eindhoven University of Technology, The
Netherlands); Rudolf H. Mak (Eindhoven University of Technology, The
Netherlands); Dick Epema (Delft University of Technology, The
Netherlands)
The proceedings of this workshop were published by Springer-Verlag
in the Lecture Notes on Computer Science series, as volume 8828.
It is also available from Springer on-line, with the option to
purchase single papers.