Ticket #2186 (closed: wontfix)

Opened 10 years ago

Last modified 5 years ago

EventNexus loading: try pre-counting events per pixel to see if this improves memory fragmentation

Reported by: Janik Zikovsky Owned by: Janik Zikovsky
Priority: major Milestone: Iteration 27
Component: Mantid Keywords:
Cc: Blocked By:
Blocking: Tester: Nick Draper

Description


Change History

comment:1 Changed 10 years ago by Janik Zikovsky

  • Status changed from new to accepted

...since it would allow the loading algorithm to reserve the event list size and not need to re-allocate the events a bunch of times.

comment:2 Changed 10 years ago by Janik Zikovsky

One might also try to estimate the mean # of events per pixel and reserve that (or a reasonable fraction of it) for all pixels.

comment:3 Changed 10 years ago by Janik Zikovsky

(In [8432]) Refs #2186: Tried precounting the # of events in each pixel before loading. Slight degradation in performance, and no clear improvement in the memory footprint growth under linux. Memory fragmentation might be improved, but an easier way to do this might be to use a custom allocator (nedmalloc maybe?)

comment:4 Changed 10 years ago by Janik Zikovsky

  • Status changed from accepted to verify
  • Resolution set to wontfix

I commented out the code for precounting in the algorithm but left it in there for possible later tests.

comment:5 Changed 10 years ago by Nick Draper

  • Status changed from verify to verifying
  • Tester set to Nick Draper

comment:6 Changed 10 years ago by Nick Draper

  • Status changed from verifying to closed

precounting does allow for some other options for memory optimization (including using boost.pool), but I'll accept this ticket for now.

comment:7 Changed 5 years ago by Stuart Campbell

This ticket has been transferred to github issue 3033

Note: See TracTickets for help on using tickets.