[Haskell-cafe] GHCi timing statistics
kumoyuki at gmail.com
Thu Mar 5 22:23:54 UTC 2015
Hi all -
Is it my imagination, or are the statistics printed by GHCi when :set +s
has been enabled entirely spurious when SMP parallelism is turned on? I'm
trying to determine if I am actually getting any improvement, and - both
by the wall clock and by the reported statistics I don't seem to be;
however, the reported statistics don't seem to line up very well with the
wall clock time.
An additional oddity that I've noted is that my algorithm's reference
implementation - which simply can't run in parallel due to a foolishly
introduced data dependency - still seems to kick 16 CPUs well into the
90%+ utilization range when I have +RTS -N. This code has no parallelism
primitives in it whatsoever.
I realize that I'm not being terribly rigorous in my description here, but
that is partly because I am not sure how to report on this yet. I think my
next step is going to be to build a standalone executable to check using
the Unix time command.
For reference I am running Haskell platform from Debian Jessie on a 16-
core amd64 system with 29GiB or RAM available. This would appear to be GHC
More information about the Haskell-Cafe