Grid Simulation with OptorSim - PowerPoint PPT Presentation

1 / 1
About This Presentation
Title:

Grid Simulation with OptorSim

Description:

... to quantify the effectiveness of the optimisation strategy under consideration. ... Among the Grid configurations currently under investigation is the LCG testbed. ... – PowerPoint PPT presentation

Number of Views:202
Avg rating:3.0/5.0
Slides: 2
Provided by: SarahP160
Category:

less

Transcript and Presenter's Notes

Title: Grid Simulation with OptorSim


1
Grid Simulation with OptorSim
Grid Replica Optimisation As the computational
and data handling requirements of large
scientific collaborations grow, Grid computing is
emerging as a feasible solution. Optimising the
use of Grid resources is crucial, and replication
of data files around the Grid is an important
tool in achieving this. With colleagues from the
European DataGrid (EDG), we have developed the
Grid simulator OptorSim to evaluate potential
optimisation strategies before they are deployed
on the real Grid.
Simplified European DataGrid Architecture
OptorSim The structure of OptorSim is based on
the EDG architecture (above). Given a Grid
topology and resources, a set of jobs to be
executed and an optimisation strategy, OptorSim
simulates how the Grid would behave under these
conditions. It provides a set of measurements
which can be used to quantify the effectiveness
of the optimisation strategy under consideration.
Optimisation Strategies There are three stages of
optimisation. First, the job should be sent to
the optimal site for running. Second, when the
job has started running, the files needed should
be made accessible as quickly as possible by
appropriate replication. Third, prediction
algorithms attempt to dynamically create replicas
that will be needed by future jobs.
Approximate LCG topology, August 2004
Results Among the Grid configurations currently
under investigation is the LCG testbed. Four job
scheduling algorithms were compared Random,
Shortest Queue, Access Cost and Queue Access
Cost. Three replication strategies were
evaluated a traditional Least Frequently Used
(LFU)-based strategy and two economic models
(binomial and Zipf), which use prediction
functions to estimate future values of files and
an auction mechanism for selecting the best
replica. Preliminary results with this
configuration show that Access Cost and Queue
Access Cost are the most efficient scheduling
algorithms. The LFU replication algorithm is
fastest in this case, but due to the large SE
sizes in this Grid, very little file deletion
occurs. Heavier loading of the resources is
required to show the true performance of the
replication algorithms, and further experiments
with the LCG testbed are ongoing.
D. Cameron, R.Carvajal-Schiaffino, P. Millar, C.
Nicholson, F. Zini
Write a Comment
User Comments (0)
About PowerShow.com