> MMAP changes nothing: as its not a sequential pattern.. and things like > asking for a term range would be extra bad.
Why's that? If the associated pages are in the system IO cache it'd be fine. On a related note, is it OK to place lets say 1K byte[]s into the Output portion of the FST? On Thu, Jun 2, 2011 at 8:36 PM, Robert Muir <rcm...@gmail.com> wrote: > On Thu, Jun 2, 2011 at 11:18 PM, Jason Rutherglen > <jason.rutherg...@gmail.com> wrote: >>> This would be very bad, if you were to lookup "foobar" you would have >>> to do a seek on every byte >> >> I think it'd be MMap'd. There was a lengthy discussion at >> LUCENE-2843, and there's an open issue to in fact store all terms in >> RAM as an FST. >> > > all in RAM is completely different. all on disk will be bad, because > when traversing terms in the FST, its random access all the way. MMAP > changes nothing: as its not a sequential pattern.. and things like > asking for a term range would be extra bad. > > --------------------------------------------------------------------- > To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org > For additional commands, e-mail: dev-h...@lucene.apache.org > > --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
