Hi! On Wed, Feb 29, 2012 at 22:21, Emmanuel Espina <espinaemman...@gmail.com> wrote: > No. But probably we can find another way to do what you want. Please > describe the problem and include some "numbers" to give us an idea of > the sizes that you are handling. Number of documents, size of the > index, etc.
Thank you! Our Solr holds currently about 168Mio documents. From each of these documents we extract the most important keywords and store them in a multivalued field (topic). Our goal is to provide faceted navigation through these topics. The topic field holds roughly 5 values per doc, but I wasn't able to compute the correct number right now. The use cases require that the facets have to be calculated fast enough so that they can be answered in reasonable time (1-2 secs) which we were able to do with a 192GB RAM machine and regular warming. Splitting the Solr into a few smaller ones (even on the same machine) seems to be the most promising way, but I've been shying away from it for some reasons: higher complexity, a huge reimport (though, I could split the current index), some components didn't support it when we were starting (Grouping was only introduced with 3.5 IIRC). I've tested the sharding approach and it was a bit slower than the one huge index approach. I'd be happy to hear some suggestions, Michael
