My thesis has been successfully defended (WOO-HOO). Fear not though, my thesis project, JOLTS, will live on.
Currently I'm adding a graphical administration tool that will allow for easier configuration of all available options. One new option being added is the ability to configure the forbidden/restricted methods. In earlier versions of JOLTS, the restricted methods were hard coded into the system. The new version will allow them to be configured as well.
The current project status can be seen on JOLTS homepage.
My thesis is concerned with investigating checkpointing and coordination issues inside grid systems. To that end, I am constructing my own grid system with checkpointing capabilities. The coordination issue will be handled by creating a special API that uses a tuple space on the central server. A prototype is currently operational. Things still left to do before I leave the prototype for the time being and move to writing my thesis proposal are:
Of course there are always lots of features that could be added to the system if time was infinite. The following is a potential list of features to add after the proposal.
implement job migration
implement worker node crash recovery
implement job crash recovery (just serialize the exception/error and return it)
check for 2 (or more) waiting ReturnResults sending different versions of the same client checkpoint, eliminate the older one
incorporate the GridProperties, so the components have some limited configuration capabilities
test the new server socket channels on the worker side
allow different machines to be configured as the server
tie in the GridInputStream class
handle serialization and reestablishing the link
convert from old sockets to socket channels
handle request for a file not in the grid (you get your exception back)
finish commenting all the classes
build a GUI front-end
have jobs leave a heavily loaded machine
the SIMD interface
the MISD interface
the tuple space
security checks to prevent invalid code calls
restrict access to specific fields
- dynamic reconfiguration of worker node attributes (not likely to happen)
- automatic update system
- remote administration of nodes?
- file writing system (not likely to be done)
- administrator tool to allow for better customization of the restricted security methods (being worked on right now)
look at serializing to memory instead of to disk deemed to be a waste of memory if the job is large
- Port the system to the .Net platform (not likely to happen any time soon)