
Jason Brazile
(180) Low Fat Grid: A RESTful Grid Service for Non-Programmers
Peer-Refereed Talk
Wednesday, 2007-06-27, 11:20 - 12:00, Arena 7
Jason Brazile - Remote Sensing Labs, University of Zurich (speaker)
Topics
Abstract
The design and implementation of a RESTful web service providing grid
functionality is presented. This web service enables small, geographically
disperse groups of non-programmers to voluntarily share their computer
resources to work on collective problems. Only three prerequisites must be
met: 1) potential compute machines must have a Java runtime environment
(JRE) installed and outgoing Internet web access, 2) at least one group
member must be able to install a single, standalone machine-independent,
unprivileged, perl CGI program on some Internet accessible server, and
3) at least one user must be able to produce an archive file representing
a batch of jobs - one directory per job, each containing a command script
and relevant input files.
The client program acts as a mini web browser: requesting jobs, executing
them, then uploading the results using the same file upload mechanism
used by web-based email clients. Job scheduling is self-balancing since
clients pull jobs rather than the server pushing them - faster clients
therefore request and process more jobs than slower clients.
A prototype of the system has been used for two scientific applications
using spare resources on heterogeneous machines at institutions located
in different countries. In one case, the application reproduced
work previously performed on super computers. In the other case,
distributed/remote execution offered a unique collaboration opportunity by
working around IP restrictions - the release-restricted software located
at one institution produced output data that could be transferred and
further processed at another institution, whose auxiliary data was
release-restricted.
In addition to its unique ease-of-deployment, performance is shown to
be favorable when compared to Condor, the leading distributed processing
framework for non-programmers.







