Hi rene, On Tue, Jul 12, 2011 at 10:12 PM, Rene Hackl-Sommer <[email protected]> wrote: > This sounds plausible, even if manually cleaning the Java cache has no > effect. Probably a JDK/JRE mismatch somewhere, just have to find the spot.
this is very likely your OS pushing frequently use data into RAM via FS cache. if you run on linux try to flush your caches with echo 3 > /proc/sys/vm/drop_caches and see if it changes anything. simon > > Thanks, > Rene > > Am 12.07.2011 19:22, schrieb Robert Muir: >> >> On Tue, Jul 12, 2011 at 10:42 AM, "René Hackl"<[email protected]> >> wrote: >>> >>> Hi, >>> >>> I am running some AutomatonQueries and see caching effects happening, >>> that I can't seem to get rid of. E.g. an AutomatonQuery takes >>> >>> Run 1: 3000 ms >>> Run 2 and all subsequent runs: 250 ms >>> >>> The IndexSearcher is closed after every full query circle (and opened via >>> open(Directory), so the IndexReader should be closed, too, right?). I have >>> changed indexes, restarted Eclipse, and run GC to no effect. I am thinking >>> FieldCache.purge might help, but how do i get there? >>> >>> Also, how to reasonably assess execution times for AQs? The very first >>> runs can last pretty long, yet cached queries are fast. Overlap seems to >>> benefit queries down the line, e.g. "*ration" is much faster if sometime >>> before "*ation" has been run. Would it be feasible to warm an index with the >>> most frequent ngrams for AutomatonQueries? >>> >> This is probably just your OS caching? this query has no caching. >> > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
