Access Issues
Can we mmap an entire Python address space into our monitor at some offset?
If not, can we use an incremental mmap technique that still gives good
performance for the kinds of accesses we need to do?
- If so, can we reuse the virtual memory hardware
The Dial Idea: turning down sensitivity for graphics updating
or for event buffering/reporting, in order to achieve performance goals.
Four Ideas for a paper
- Port Alamo to Python
- Measuring the amount of indirection (and other redundant calculations)
in the virtual machine, relative to the actual actual productive work
getting done. (like heat loss in an energy transformation)
- Dynamic Profiler - GUI, dials, watches and reports at multiple
granularities of time steps
- Redo Python so it can be easily multiprogrammed and/or is properly
reentrant and thread-safe yet amazingly efficient.
Even though the Python Software Foundation grant may not loom large
on Rincon's radar, participating in that community and getting its
sponsorship might add credibility with other funding agencies.
Rich is working on C instrumentation on Monday. Meeting with Mike,
Rich, and Clint on Tuesday afternoon, tba.
We hope to get some additional events (such as list or memory events)
instrumented and new visualizations working before Rich leaves.