On 7/30/07, David Whalen <[EMAIL PROTECTED]> wrote: > Hi All. > > I am using facets to help me build an ajax-driven tree for > search results. When the search is first run, all I need to > do is show the counts per facet, for example > > search results for "fred" > +--A (102) > +--B (234) > +--C (721) > +--D (512) > > sounds simple, but I also need to break-down the results from > "D" by a different index in lucene: > > search results for "fred" > +--A (102) > +--B (234) > +--C (721) > +--D (512) > +--D1 (19) > +--D2 (34) > +--D3 (45) > > what I have been doing in my solr querystring looks like this: > > rows=0&facet=true&facet.limit=-1&facet.field=<field1>&facet.field=<field > 2> > > Unfortunately we're seeing really bad performance and we're > constantly running out of heap space on this type of query. > > So, my question is, would breaking this into two calls perform > better? That is, > > rows=0&facet=true&facet.limit=-1&facet.field=<field1> > > and then > > rows=0&facet=true&facet.limit=-1&facet.field=<field2> > > ? > > It seems to me that two calls would have more overhead than one, > but it might lessen the impact on the heap space on my server.
The memory needs would be the same, but with more CPU overhead for 2 requests. If you can't increase the heap enough, or you can't reduce the memory via upgrading to Solr 1.2 and using facet.enum.cache.minDf, then what you could do is have two search servers (or two sets of search servers) and send all facet requests for field1 to server1 and all facet requests for field2 to server2. -Yonik
