I would suggest you force your software to use only one processor on both dual- and quad-core machines and check the timings again. If the timings are in the range of expectancy (due to hardware speed/processor frequency) then your software has a bug related to NUMA. The non-uniform memory architecture used by AMD since the Athlon64/Opteron line penalizes non-local memory access - especially in an SMP scenario.
Mike > -----Ursprüngliche Nachricht----- > Von: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] Im Auftrag von > Benilton Carvalho > Gesendet: Montag, 8. September 2008 21:20 > An: sqlite-users@sqlite.org > Betreff: [sqlite] SQLite vs. quad-cores > > Hi everyone, > > I'm a BioConductor developer and we use SQLite (via RSQLite > package) to handle annotation of high-density oligonucleotide > microarrays. > > Our solution, implemented in the pdInfoBuilder package, > worked great until we got machines with quad-core processors. > One particular setup is a machine with 2 Quad-Core AMD > Opteron Processor 2356. The processes that usually took > 1-2 hours on older machines (dual-cores AMD), now take 6-8 > hours on the new computers. > > We thought there would be something wrong with our switch > (given the particular network scenario we have here), and > changing it didn't solve the problem. Later, we decided to > run everything on local storage (instead of network), and the > same behavior was observed again. > > Is there anything else that you guys could recommend? > > thanks a bunch, > > benilton > _______________________________________________ > sqlite-users mailing list > sqlite-users@sqlite.org > http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users > _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
