search

bblum / landslide-simics

the landslide codebase (old simics version)
11 stars 9 forks source link

estimates: RE behaves strangely when WBE says exactly 50% #67

Open bblum opened 9 years ago

bblum commented 9 years ago

note huge drop in RE-based time estimate (microseconds) vs WBE-based branch/percentage estimate appearing normal. what's going on?

[EXPLORE]       from #19/tid2, chose tid 2, child of #0/tid-1
[ESTIMATE]      current total usecs 48428884, prop 50.000000%, est total 96856444
[LANDSLIDE]     End of branch #1040.
[LANDSLIDE]     Estimate: 50.000000%
...
...contents of one branch omitted...
...
[EXPLORE]       from #18/tid2, chose tid 2, child of #16/tid3
[ESTIMATE]      current total usecs 49175687, prop 50.390625%, est total 52142906
[LANDSLIDE]     End of branch #1041.
[LANDSLIDE]     Estimate: 50.390625%
...
...many branches omitted...
...
[EXPLORE]       found no tagged siblings on current branch!
[ESTIMATE]      current total secs 53807330, prop 100.000000%, est total 53806616
[LANDSLIDE]     End of branch #1140.
[LANDSLIDE]     Estimate: 100.000000%
...

note that WBE's guess of 50% is also quite wrong.

seen so far only on test config:

KERNEL_IMG=kernel-pobbles
TEST_CASE=yield_vanish
within_function vanish
within_function thread_fork
within_function yield
without_function wait_reap
bblum commented 9 years ago

still working on debugging tiny discrepancy between 100%-time perfect-information small discrepancy between estimated total and actual total. strange behaviour may be an artifact of whatever is causing that too.