Hi all,
I summarized the benchmarking comparison. Here's what I did:
* collected all matrix, study and tree IDs
* wrote a perl script that fetches the PhyloWS URL for each data type,
for each ID, for each server and recorded response time
* wrote out the results in the attached files (e.g MATRICES.sdsc lists
the response time in seconds for each matrix when downloaded from
8ball.sdsc.edu)
* joined the equivalent files for both servers (e.g. $ join
MATRICES.sdsc MATRICES.nescent > MATRICES.joined)
* created box plots, which are the attached pdfs
Conclusions: for pages where the main effort lies in tomcat returning
JSPs (i.e. for studies and trees), NESCent outperforms SDSC (this is
after I assigned more memory to tomcat). For pages where the main
effort lies in the database returning data, SDSC (DB2) outperforms
NESCent (postgres) by about 30%. The low hanging fruit for further
optimization seems to be the database, not the servlet container.
I consider this item on the release plan as 100% complete (green) and
will now close issue #2899241
Rutger
--
Dr. Rutger A. Vos
School of Biological Sciences
Philip Lyle Building, Level 4
University of Reading
Reading
RG6 6BX
United Kingdom
Tel: +44 (0) 118 378 7535
http://www.nexml.org
http://rutgervos.blogspot.com
|
