GridPP

1
Dynamic Grid Optimisation
Grid Optimisation Intelligent middleware is key
to the operation of large-scale data grids. Along
with collaborators from the European DataGrid
1, we are working to construct middleware for
the petabyte-scale UK Grid for Particle Physics.
Final middleware products must manage efficiently
use of resources the network bandwidth, storage
capacity and processing power of the Grid.
Minimisation of network loading and maximisation
of storage usage is possible by implementation of
dynamic file replica management.
European DataGrid Architecture
Grid Simulation Within the European DataGrid
architectural design, a replica manager service
is present at every site, providing optimised
file access via an internal replica optimiser.
The role of the replica optimiser is twofold it
must provide the most efficient access to data
for currently executing jobs, and through dynamic
replica creation reduce data-access latencies for
all users of the Grid. The need for testing
possible replica optimisation algorithms has led
to the construction of a Grid simulation called
OptorSim 2. OptorSim provides an artificial
data grid infrastructure to allow testing of many
different replication strategies. A developer
using the software is able to describe
site-to-site network connections, together with
individual site resources. Further than this the
simulation provides a framework for a developer
to describe site policies and job descriptions,
where jobs are described by a list of associated
files. At simulation runtime these jobs are
distributed via a Resource Broker to Computing
Elements within the Grid.
Simulation of GridPP Testbed
Results and Future Work Some results from the
simulation appear in 4 where the European
DataGrids testbed was simulated. Since there are
no large-scale physics experiments currently
running at CERN, tests were performed using data
on file sets being used at the Collider Detector
at Fermilab. The results show the overall time to
simulate 10000 Grid jobs, for each of the four
replication strategies with four different access
patterns describing the order in which files are
read by the simulated job. From the results it is
clear that the Economic Model is the optimal
solution for sequential access and performs well
under the other access pattern conditions. Work
is currently underway to simulate a Vickery
auction process within the Economic Model
optimisation algorithm. It is expected that
addition of this auction will lead to a more
stable and smooth optimisation state. Recent
work has focussed on making the simulation more
realistic for the GridPP testbed, incorporating
background network traffic and simulating the
particle physics user environment.

• Replication strategies
• no replication
• always replicate, delete oldest file
• always replicate, delete least accessed file
• economic model

1 The DataGrid Project. http//www.eu-datagrid.o
rg/ 2 OptorSim A Replica Optimisation
Simulation http//cern.ch/grid-data-management/opt
imisation/optor/ 3 L. Capozza, K. Stockinger,
and F. Zini. Preliminary Evaluation of Revenue
Prediction Functions for Economically-Effective
File Replication, June 2002 4 W. H. Bell, D. G.
Cameron, L. Capozza, A. P. Millar, K. Stockinger,
F. Zini. Simulation of Dynamic Grid Replication
Strategies in OptorSim, in Grid 2002, November
2002
Results from 4 showing total job time for 10000
simulated jobs.