KORE
09/29/2005
Bala is setup to work on UofT machines.
Current Work:
- 1 Fixing the buffering issues between traceServer and the pem client in KORE. 2 Filling the gaps in the memory map.
I guess once we have (1) running, we can do an overhaul on KORE to have the delta updates.
We will be meeting soon to discuss and set up what we do next.
10/06/2005
Meeting: Jonathan and Bala
- Bala is working on the buffering issues between the oisServer and the buffer client
- Bala will upload stuff he did for the poster on to this page as well as a paper outline he has been working on.
- I will review and work on the paper that Bala uploads we are targeting a 5 - 10 page paper.
- We will try and arrange a time to give the demo again to Orran.
10/06/2005
Attached files kore.tex (report outline) and kore.ppt
10/08/2005
Fixed the buffering problem by sending the data in 4K chunks.
10/19/2005
Current status:
- PMRoot is included in the map of physical memory. How to include FCMStartup?
The following are fields from FCMStartup <root, imageOffset, imageSize> 0x8002000000241e80 0x5f00000 0xdd58 0x80020000002419c0 0x2688000 0x70f000 0x8002000000242cc0 0x2ea0000 0x1000 0x8002000000242700 0x2688000 0x70f000 0x8002000000246e00 0x2688000 0x70f000 0x8002000000248980 0x2688000 0x70f000
- PEM trace collection working fine - some irregularities with types of alloc'ed clustered objects observed:
The following is a sample trace. An Alloc event has <Type (or) gte if type unavailable, misshandler, lte>. A Dealloc event is of type <lte, gte> Alloc CObjRoot 0x8002000000259cc0 0x8000000010003960 Alloc 0x000010008004f308 0x0000100006641800 0x0000010010000180 Alloc CObjRoot 0x800200000025e380 0x8000000010003640 Alloc CObjRoot 0x800200000025f540 0x80000000100038f0 Alloc CObjRoot 0x800200000025be80 0x8000000010003bd0 Alloc CObjRoot 0x8002000000256ec0 0x80000000100039c0 Alloc 0x000010008004f308 0x0000100006641340 0x0000010010000190 Alloc 0x000010008004f308 0x0000100006641280 0x00000100100001a0 Dealloc 0x8000000010003960 0x8000000000003960 Dealloc 0x8000000010003bd0 0x8000000000003bd0 Dealloc 0x8000000010003640 0x8000000000003640 Dealloc 0x80000000100038f0 0x80000000000038f0 Dealloc 0x80000000100039c0 0x80000000000039c0 Dealloc 0x0000010010000180 0x0000010000000180 Dealloc 0x0000010010000190 0x0000010000000190
10/28/2005
11/04/2005
PEM trace collection is working, no known problems. Passed a few sanity checks, consistent with list of objects got from oisServer. Next task is measurements.
11/11/2005
Measurements to be done (Jonathan please edit this)
Overheads of an OIS interaction
- Time spent at Tcl
- Time spent on the wire
- Time spent at the server's side
- Collecting the list of objects to be sent (for printObjs)
Overheads of a GDB interaction
- Time spent at Tcl
- Time spent at client-side GDB
- Time spent on the wire
- Time spent at server's side
- Time spent at the GDB stub
Overheads of tracing
- Same as OIS interaction, but include
- time to collect trace buffers
- time to interpret trace
11/22/2005
Above list looks good to me.
12/12/2005
Kore patch (mainly KORE's pem stuff) k42KorePatch
Incomplete (just added timing calls) versions for measuring overheads: ois.C and gdb.tcl