#4225: RTS -i (heap profiling interval) documentation could be clearer
---------------------------------+------------------------------------------
Reporter: ClaudiusMaximus | Owner:
Type: bug | Status: new
Priority: high | Milestone: 6.14.1
Component: Profiling | Version: 6.12.3
Keywords: | Testcase:
Blockedby: | Difficulty:
Os: Unknown/Multiple | Blocking:
Architecture: Unknown/Multiple | Failure: Documentation bug
---------------------------------+------------------------------------------
Comment(by simonmar):
To clarify/confirm the points that have come up so far:
* heap census time is not counted as runtime in the profile
* GC time ''is'' counted as runtime
* heap profiling triggers a full GC at intervals specified by the `-i`
flag, where the default is 0.1s.
* There is no way, currently, to say "do a census after every naturally
occurring GC". That could be easily implemented, although in GHC it would
be less useful than it was in hbc/nhc98, because GHC has a generational GC
and full collections are quite infrequent - you would get very few
samples.
Eliminating profiling overhead from runtime isn't really possible anyway,
because other aspects of profiling (e.g. maintaining the cost centre
stacks) add quite a bit of overhead too, and that can't be easily factored
out. Perhaps not counting the time for the "extra" GCs as part of the
runtime would help in that different profiles would be more easily
comparable, though it still won't help when comparing profiling to non-
profiling runs.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/4225#comment:7>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
