The Developer Perspective The Application Hosting Environment Stefan Zasada Centre for Computational Science

advertisement
The Developer Perspective
The Application Hosting Environment
Stefan Zasada
Centre for Computational Science
University College London
Introduction
• User requirements/motivations
• Identifying and understanding requirements
• The design of the AHE
• User interaction
• Why we took this approach
• Development issues
• Future plans
User Requirements
• Quickly and easily run applications on remote
resources
• Move input files from desktop to grid resource
• Stage files back after execution
• Steer simulation as it’s running
• Visualize results
Motivations
• Quickly and easily run scientific applications
on the grid, e.g. molecular dynamics codes
such as NAMD
• Very often a group of users will want to access
the same application
• Install the application once, then provide a
service to allow others to share it
• Hide complexity of submitting jobs to grid
Design
•
Design arisen from previous experience using
grid middleware:
–
–
–
–
Difficult to configure and/or install
Dependant on lots of supporting software
Require modified versions of common libraries
Require lots of non-standard ports to be opened
on firewall
– Large footprint – memory/disk space
•
Base on ideas of WEDS
Design
• SOA Approach - WRSF Compliant
• Applications represented as stateful WS-Resources
• User interacts with the WS-Resource to manage
application
• Uses WSRF::Lite Perl toolkit
• Uses GridSAM for job submission
• MyProxy - required by GridSAM when submitting to
Globus
Architecture of the AHE
AHE Client Interaction
• Java command line and GUI clients provided
• Clients allow user to prepare application,
stage files and submit
• GUI client implemented a wizard that takes
user through steps of starting application
• Client parses app config file to find files to
stage in and out
• Parser plug-ins can be written for new apps
Why we took this approach
• We provide a programmatic interface to manage
running application
• Allows a variety of different clients to be developed
• Running client on local machine allows us to stage
files more easily
• Less restrictive that forcing users to use a portal
• Some apps require local visualization clients to be
installed
Benefits of the design
• Command line clients can easily be called from script
to automate complex workflow
• When investigating mutation patterns of HIV-1
protease we need to perform equilibration protocols
that involve chained MD simulations
• Easy to administer:
– server can be deployed be expert user on workgroup by
workgroup basis
– Client can be deployed by user
Technical Strategy
• Standards chosen: WSRF, MyProxy, WebDav
SSL/X.509
• Services provided dictated by GridSAM
• Operations necessary to manage application:
prepare, stage files, start, terminate, destroy
• GridSAM offers a common set of operations for
all DRMs supported
Development Issues
• Time: minimum set of functions available in
first release; plan to add more
– Would have been nice to have more time for
evaluation
• Java/Perl web services interoperate pretty well
• Too much input from users?
– Everyone has different requirements they want
supporting
Evaluation
• Difficult to establish evaluation metrics
– How long it takes to submit a job is dependant on
the job type
• Does it make using the grid simpler - yes
• Real world evaluation - used to run production
jobs
• Need to plan in more time for rigorous
evaluation into future release cycles
Lessons Learnt
• Most successful:
– AHE used for production runs in the NGS and
TeraGrid
– Simple design and command line clients allow for
complex scientific workflows to be constructed
• Least successful:
– Implementation needed some modification to run
in Tomcat
– Too early to tell?
Future Plans
• Integrate into OMII stack
• Orchestrate complex workflows (using BPEL,
Taverna?)
• Use RealityGrid steering web service
• Coupled models – host applications which are
made up of other application components
• Clients to run on a PDA/mobile phone
Summary
• AHE is a kind of portal
– Platform on which we can build
• Lightweight and useable
• Providing programmatic interface allows different
types of client to be developed that meet different
user needs
• Allows for scientific applications to be easily
deployed and shared
• Lightweight nature requires minimum input from
system administrators etc.
Acknowledgements
• UCL: Peter Coveney, Matt Harvey, Laurent
Pedesseau, Radhika Saksena, James Suter,
Phil Fowler, Kashif Sadiq, Mary-Ann Thyveetil,
Giovanni Giupponni, Simon Clifford
• Manchester: Mark McKeown, Stephen Pickles,
Rob Haines, Andy Porter
• EPSRC
• OMII
Links
• RealityGrid web site:
http://www.realitygrid.org/AHE
• NeSCForge:
http://forge.nesc.ac.uk/projects/ahe/
• Mailing list:
http://www.mailinglists.ucl.ac.uk/mailman/listinfo/ahe-discuss
Download