Information Technology Reference
In-Depth Information
Box 1.
Executable = “AxWorker.pl”;
Arguments = “-j ax-May1319-41-28 -p 100 -1 2048 -2 2048 -3 2025 -4 2031 \
-A gsiftp://example.org/dpm/home/biomed/heinz/selection_2048.mat-ax-
May1319-41-28 \
-B gsiftp://example.org/dpm/home/biomed/heinz/selection_2048_P.pco-ax-
May1319-41-28 \
-C gsiftp://example.org/dpm/home/biomed/heinz/selection_2048_H.tra-ax-
May1319-41-28 -i 1”;
Stdoutput = “output.txt”;
InputSandbox = {“/home/stockinger/AxWorker.pl”, “/home/stockinger/AxParafitB-
LAS”};
OutputSandbox = {“output.txt”}
EXPERIMENTAL RESULTS
which means that experimental results can not be
fully reproduced. However, once one is correctly
registered with the Virtual Organization, one can
use it any time of the day.
On the client side, we used gLite on GNU/
Linux on an AMD Opteron machine (2 GB RAM,
2.2 GHz CPU) located in Lyon, France - previous
tests (in particular with the installation of CopyCat
and the Grid interface have been conducted on a
machine located in Lausanne, Switzerland). The
gLite components used are the workload man-
agement system (for job submission and status
monitoring) as well as data management clients
for file transfer. Additionally, we deployed and
used a Task Server that is located in Lausanne,
using resources provided by the Vital-IT group
of the Swiss Institute of Bioinformatics. In the
second experiment, we used a dedicated compute
cluster with 128 CPUs. In contrast to the Grid, the
cluster had to be reserved in advance.
The main goal of the gridified version of
CopyCat(AxParafit) is to accelerate and facilitate
large-scale analyses. We present two experiments
with large computational demands and study
their performance on the Grid. The performance
improvement is outlined with respect to running
the application sequentially on a single machine.
Moreover, we conduct a performance comparison
between a dedicated compute cluster and the Grid.
Test Environment
The Grid platform that is supported by our ap-
plication is gLite 3. Tests are conducted using
gLite on the EGEE production infrastructure. In
particular, we use the Virtual Organization (VO)
that is dedicated to biomedical applications:
“biomed”. Members of this VO have access to
about 50 Computing Elements (acting as front-
ends to computing clusters), each having between
2 and a few hundred processing cores. The exact
number of processing cores available to a single
user at a given time cannot be easily obtained since
it depends on the current system load as well as
the general availability of a Computing Element
at a certain point in time. Currently, gLite does
not support resource reservation nor job priorities,
Experiment with Real-world Data
In the first experiment we are interested in the
raw performance (response time) of AxParafit.pl,
i.e., how long does it take to fully process a set of
tasks on the Grid. In this experiment, we do not
include CopyCat but directly invoke AxParafit.
pl as shown in Box 2.
Search WWH ::
Custom Search