From: Neal R. <ne...@ri...> - 2005-07-26 16:32:38
|
On Tue, 26 Jul 2005, Geoffrey Hutchison wrote: > > On Jul 26, 2005, at 9:59 AM, Gustave T. Stresen-Reuter wrote: > > > One way to improve the speed of htdig on mac os x is to take > > advantage of the AltiVec processor. I don't know anything about how > > to do this, but in reading some documentation I ran across a > > document says that "Existing C code written for serial execution" > > can take advantage of the AltiVec processor. > > I'm not sure there's really a lot that can happen here. Neal can > contradict me, but the last time I did benchmarking/profiling on the > code, most of the slowdown during indexing was in the database code, > and a large part of that was I/O bound. What is your wordlist_cache_size set to? Make sure that it is about 2-3% of the expected index size. I would also disable all index compression and run htdig again to see if you notice a speed up. Setting wordlist_compress & wordlist_compress_zlib to 'false' is the way to test that. > Furthermore, AltiVec really shines doing floating-point processing, > e.g. matrix multiplication. There's very, very little of that in ht:// > Dig. > > I think for future development (i.e., 4.0) multi-threading the > indexing would probably help *MUCH* more than adding AltiVec or SSE > processor-specific optimizations which mostly help with floating- > point ops. > > And of course, moving to a different database backend would probably > help immensely too. Yep. -- Neal Richter Knowledgebase Developer RightNow Technologies, Inc. Customer Service for Every Web Site Office: 406-522-1485 |