On Thu, Jun 23, 2011 at 8:27 AM, Johan Tibell <[email protected]> wrote: >> Is 5 the optimal number of bits to slice off at a time (ie the best >> fanout)? It sounds like node copy cost on insert argues for a >> slightly narrower fanout. You'll be evacuating / scanning more words >> total, but new nodes may equate to less scanning overall (especially >> if this is running long enough to have some nodes get tenure). > > I'm experimenting with this. 6 is far too much, making inserts 4-5x > slower. 4 doesn't seem to improve things much (which is a bit > counter-intuitive given that 6 made things so much work), but I need > to experiment some more.
After some more testing I can improve the performance of insert by 30% by reducing the array size from 32 to 16. GC still dominates though. Johan _______________________________________________ Glasgow-haskell-users mailing list [email protected] http://www.haskell.org/mailman/listinfo/glasgow-haskell-users
