Marc, What do you mean by Katta's ranking algorithm? If you use SOLR-1395's search request system that traverses Hadoop RPC, it's simply using what Solr offers today in terms of distributed search (i.e. no distributed IDF). Instead of requests being serialized into an HTTP call, they are serialized over Hadoop RPC. That's the only difference. Katta's main benefit to me isn't it's integration with search, but it's management of shard/cores over a large cluster. This is a major benefit to Solr, where today, if a server goes down or slows down (which is worse), all queries start to back up and fail, and the shard proxy server needs to be restarted because dead threads have exceeded the app server's thread pool limit. This should not happen on production servers. Also Solr needs the ability to return partial results if a core is unavailable, this will be much easier to implement with SOLR-1395 because the dead node may be marked as such in Zookeeper and future requests sent to an available replicated shard.
It should be fairly easy to add to Katta (if it's not there already) the ability to send a request to the shard/core with the lowest load. There's also recently been work in adding new shard distribution policies. -J On Fri, Oct 9, 2009 at 4:03 AM, Marc Sturlese <marc.sturl...@gmail.com> wrote: > > Hey there, > I am trying to set up the Katta integration plugin. I would like to know if > Katta's ranking algorith is used when searching among shards. In case yes, > would it mean it solves the problem with IDF's of distributed Solr? > -- > View this message in context: > http://www.nabble.com/SOLR-1395-integration-with-katta.-Question-about-Katta%27s-ranking-among-shards-and-IDF%27s-tp25819241p25819241.html > Sent from the Solr - User mailing list archive at Nabble.com. > >
