Next: Server Architecture
Up: Analysis Server
Previous: Analysis Server
During the evolution of the Vampir project, we identified a set of abstract
requirements with respect to current cluster platforms that typically cannot
be fulfilled by most classical sequential post mortem software analysis
approaches:
- Do calculations in parallel.
- Support distributed memory.
- Increase scalability for both long (regarding time) and wide
(regarding number of processes) program traces.
- Allow preliminary cancellation of requests. After the user spawned
an analysis process, it should be possible to terminate the process without
waiting for its completion as it might turn out to be very time consuming.
With regards to a parallel infrastructure, this requirement is not easy to
fulfill.
- Limit the data transferred to the visualization client to a volume
that is independent of the amount of traced event data.
- Portability. The analysis component should work on as many
parallel platforms as possible.
- Extensibility. Depending on user demands, new analysis
capabilities should be easy to add.
2003-10-06