For locus-0.87, I got a significant speedup by increasing the size of memory cache: Decline and Fall of the Roman Empire now indexes in 15 minutes. (With only 3 indexing cycles - I should probably pick up some bigger reference collection.)


Performance measurably (although probably not noticeably) improved for locus-0.8, compiled with gcc 2.8.1. Time to index 9 772 883 bytes of Gibbon's Decline and Fall of the Roman Empire is now 26 minutes.


locus-0.64 (and later) - compiled with gcc 2.8.0, which can compile my convoluted C++ with optimizations enabled (as opposed to gcc 2.7.2.3) - is a definite improvement over the previous version. Time to index 9 772 883 bytes of Gibbon's Decline and Fall of the Roman Empire went from 62 to 29 minutes.

Time/size dependency is pretty linear. I guess it stays linear till you have RAM, but surely not forever: 49 752 228 bytes of VIRUS-L archive takes almost 6 hours (5 hr 43 min, to be precise). Here, disk writes are the bottleneck (stop list cuts the time to 4 hr 45 min).


Of course, many more tests should be done, but I'm too lazy. If you do your own measurements, I'll gladly include them here.
Top: locus homepage