|
Delft University of Technology (TUD) participates in the following scientific CoreGrid activities:
- WP6: Institute on Resource Management and Scheduling
- Task 6.1: Definition of the Components of a GRID Scheduling Architecture and their Interaction
- Task 6.2: Multi-Level Scheduling Strategies
- Task 6.5: Evaluation and Benchmarking of GRID Scheduling Systems
- Task 6.7: Coordinating GRID Scheduling with Data Management (Task leader)
We are interested in the design choices for hierarchical and
peer-to-peer-like solutions for grid schedulers,
and in their actual implementation and testing.
We have defined a scheduling architecture for such a scheduler
in multiclusters, and actually implemented it
on our DAS system (the KOALA scheduler).
We have designed strategies for processor co-allocation
in multiclusters, which is in effect two-level scheduling,
and we have implemented support for co-allocation in our
KOALA scheduler.
| Participants: |
Dick Epema |
|
Hashim Mohamed |
|
Ozan Sonmez |
We have assessed the performance of strategies for scheduling
and co-allocation in multiclusters, both with simulations
and with measurements of KOALA on our DAS testbed.
We intend to extend this work to different workloads,
and to do much more elaborate performance tests.
KOALA includes support for moving (large)
input files to the location(s) where applications will
run, which are selected based on estimated file transfer
times.
We intend to extend KOALA to datamining applications, and to
compute- and data-intensive coputational-steering applications
with a simulation and a visualization component.
TUD has, or has planned, the followijg collaborations in CoreGRID:
-
An MSc student from CNRS (Frederic Desprez/ENS-LYON), Alexis Ballier, is visiting
TUD over the summer of 2005 to perform simulations of reservations
and co-allocation in grids.
-
Together with the University of Muenster (Sergei Gorlatch), we will have a
CoreGRID fellow on the topic of
Scheduling algorithms for higher-order components on the grid
(start november 2005).
| Participate in CoreGrid Workshop in Pisa, january 13, 2005 (Dick Epema) |
done |
| Visit to ENS-Lyon, may 23 and 24, 2005 (Dick Epema and Hashim Mohamed) |
done |
| Visit from the University of Manchester (Rizos Sakellariou) may 27, 2005 |
done |
| Participate in the CoreGrid WP6 Workshop in Barcelona, june 13 and 14, 2005 (Dick Epema and Hashim Mohamed) |
done |
| Attending Europar and the CoreGrid MGA in Lissabon, aug. 30 - 2 sept., 2005 (Dick Epema) |
done |
| Creating a first release of the KOALA co-allocating scheduler on the DAS (sept. 2005) |
done |
| Visit from the University of Dortmund (Ramin Yahyapour and Carsten Ernemann) oct. 27-28, 2005 |
done |
| Visit to the University of Dortmund, nov. 2005 (Alexandru Iosup) |
done |
| Attend the CoreGRID Integration Workshop in Pisa, nov. 28-30, 2005 (Alexandru Iosup) |
done |
| Stay of Jeremy Buisson (INRIA) in Delft in the REP, january-may 2007 |
done |
| Attending the WP6 meeting, Manchester, may 2007 (Alexendru Iosup and Hashim Mohamed) |
done |
| Visit to the University of Chile, may 2007 (Alexandru Iosup) |
done |
| Attending the CoreGRID Symposium, Europar, and the WP6 meeting, Rennes, august 2007 (Dick Epema and Ozan Sonmez) |
done |
| Creating a workload generator and benchmarking suite for grids |
in progress |
| Extending KOALA to support different application types |
in progress |
Priority queues
Multiprocessor scheduling policies
Condor Flocking
Simulations of co-allocation
The KOALA co-allocating scheduler
The main objectives of TUD with respect to grids is to design, implement,
and test grid schedulers, and to add to the understanding
of fundamental scheduling and resource management issues in grids.
T6.1: Grid Scheduling Architecture
- Here we want to contribute to the definition of grid scheduling archirectures.
T6.2: Multi-Level Scheduling Strategies
- Research co-allocation.
T6.5: Evaluation and Benchmarking of GRID Scheduling Systems
- Do both performance tests with KOALA, and perform simulations of
fundamental problems in grid scheduling.
T6.7: Coordinating GRID Scheduling with Data Management
- Adapt KOALA to deal with different types of data-intensive jobs.
Below, we have collected our most important papers and presentations
with respect to our work in WP6.
For more, see this page.
-
D.H.J. Epema, Grid Research at TU Delft,
CoreGRID Scientific Advisory Board Meeting, London, 28 february 2007.
presentation ![Presentation, PPT [2030KB]](../icos/icoppt.gif)
-
H.H. Mohamed
and D.H.J. Epema, Experiences with the KOALA Co-Allocating
Scheduler in Multiclusters, Proc. of the 5th IEEE/ACM
Int'l Symp. on Cluster Computing and the GRID (CCGrid2005),
Cardiff, May 2005.
article ![Article, PDF [175KB]](../icos/icopdf.gif)
-
H.H. Mohamed
and D.H.J. Epema, An Evaluation of the Close-to-Files
Processor and Data Co-Allocation Policy in Multiclusters,
CLUSTER 2004, IEEE Int'l Conference Cluster Computing 2004,
September 2004.
article
-
S. Banen,
A.I.D. Bucur, and D.H.J. Epema, A Measurement-Based
Simulation Study of Processor Co-Allocation in Multicluster
Systems, Ninth Workshop on Job Scheduling Strategies for
Parallel Processing (in conjunction with HPDC-12), D.G. Feitelson,
L. Rudolph and U. Schwiegelshohn (eds), Seattle, USA, june
2003, LNCS 2862, 105-128, 2003.
article
-
A.I.D.
Bucur and D.H.J. Epema, Trace-Based Simulations
of Processor Co-Allocation Policies in Multiclusters,
12th IEEE Int'l Symp. on High Performance Distributed Computing
(HPDC-12), Seattle, Wa, USA, june 2003, 70-79, 2003.
article
-
The KOALA
Grid Scheduler: Architecture and Issues, WP6 Workshop, Barcelona,
june 13-14,
2005.
presentation
-
KOALA: A Co-Allocating
Grid Scheduler, VL-E Workshop, Amsterdam, april
5, 2005.
presentation
-
KOALA:
Processor and Data Co-Allocation in Grids, the Dagstuhl
seminar on Future Generation Grids, 1-5 nov. 2004.
presentation
The KOALA Co-allocating Grid Scheduler
The Parallel and Distributed Systems (PDS) Group in Delft
How to reach PDS/TUD
Delft University of Technology
The Dutch Virtual Laboratory for e-Science Project
The Distributed ASCI Supercomputer (DAS)
|
