Visualisierung auf einem Grid-System mittels verteiltem Java AWT

Student:Bernd Heller
Title:Visualisierung auf einem Grid-System mittels verteiltem Java AWT
Type:diploma thesis
Advisors:Veldema, R.; Philippsen, M.
State:submitted on May 23, 2005

This DA is to design and develop a grid-enabled Java Abstract Windowing Toolkit (AWT) for our research Grid-DSM Jackal. In short, our Jackal system allows a programmer to automatically distributed a multithreaded (thus already parallelized) program over a number of machines connected by some interconnect. Our goal is then to provide a Single System Image (SSI). As part of a SSI, it should provide a unified view of AWT as Java normally provides.

The problem is then that, one machine might create a 'window' and give the 'handle' to the window to another machine (by for example storing the 'handle' in an object where the DSM protocol has transfered the object to another machine) where the 'handle' is normally only locally valid.

This DA should then 'virtualize' the window 'handle' so that each machine can create AWT window 'handles' that should then be globally valid.

The second problem is then that our Jackal system targets the 'grid', which loosely translates to 'clusters connected by the internet'. This means that, if a machine were to put single pixels on a display that is in terms of latency, far away, there is a performance problem. This problem could be solved in a number of ways: for example buffering the pixels locally for some period of time, compressing the pixel stream if needed, reducing bitmap resolutions if the programmer allows it (lossless and lossy compression). This DA should investigate which solutions can be automatically applied (which is different than current works which need programmar input always).

Another problem is event management, for example, 'draw line', 'mouse click, 'mouse move' and 'update graphics' events. These events should be delivered in order and some events can be ignored if performance is critical. For example, if a mouse delivers 1000 'move' events to some machine far away, we could throw away a large number of these except the last. Another aspect of this DA would be to investigate mechanisms to optimize event management.

watermark seal