Previously, I replied to another thread on Vectors. The attached .xsl file which shows avg. msec. for changing buffer sizes, reported from an experiment on TREC collection.
This might help for a decision. Cheers, Murat > I recall somebody (Doug, I think, but I could be wrong) trying different > buffer sizes and reporting insignificant performance changes. But please > provide a patch in the JIRA issue, and people can vote for it, if they > like it. > > Otis > > ----- Original Message ---- > From: Tatu Saloranta <[EMAIL PROTECTED]> > To: java-dev@lucene.apache.org; Maxim Patramanskij <[EMAIL PROTECTED]> > Sent: Saturday, May 6, 2006 7:11:42 PM > Subject: Re: 2.0 release > > --- Maxim Patramanskij <[EMAIL PROTECTED]> wrote: > >> Currently, buffer sizes for BufferedIndexInput and >> BufferedIndexOutput are equals and have constant >> size of 1024 bytes. >> >> When using a database for index persistence, it >> slowdowns performance much >> because of relatively small buffer size. With >> JDBCDirectory and buffer >> size increased from 1Kb to 16Kb I got 200%(3 times) >> performance >> increase for indexing. >> >> What about adding a possibility to change buffer >> size on runtime, that >> client code can set appropriate values depending on >> what persistence >> type for indexing is used? > > Adding configurability is good, and also changing the > default value to be bit more optimal would be even > better? Just changing default to 2k or 4k (4k is close > to a physical page size, might be good default?) might > be worth doing as well. > I mean, 90% of users probably use things exactly as > they are configured by default; so this would probably > make many Lucene users happy. ;-) > > -+ Tatu +- > > > __________________________________________________ > Do You Yahoo!? > Tired of spam? Yahoo! Mail has the best spam protection around > http://mail.yahoo.com > > --------------------------------------------------------------------- > 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] > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
