I'd say you are at a point where sharding may well help. But, as others
have suggested, you have other issues to consider first - less memory
for Solr and upgrade to a more modern Solr. 

Also, if as Yonik asks only the first query is slow, you can set up a
NewSearcher query in your solrconfig.xml to run this first query on
every commit, meaning your users will always get faster queries.

Upayavira

On Tue, Nov 13, 2012, at 11:16 AM, Aeroox Aeroox wrote:
> Thanks  Yonik.
> 
> Should I consider sharding in this case ( actually I have one big index
> with replication) ? Or create 2 index (one for search and other for facet
> on a different machine) ?
> 
> Thanks folks
> 
> With love from Paris (it's raining today :(
> 
> Le mardi 13 novembre 2012, Yonik Seeley a écrit :
> 
> > On Mon, Nov 12, 2012 at 8:39 PM, Aeroox Aeroox 
> > <aero...@gmail.com<javascript:;>>
> > wrote:
> > > Hi folks,
> > >
> > > I have a solr index with up to 50M documents. A document contain 62
> > fields
> > > (docid, name, location....).
> > >
> > > The facet count took 1 to 2 minutes with this params :
> > >
> > > http://XXXX.../select/?q=solr&;
> > >
> > version=2.2&start=0&rows=0&facet=true&facet.limit=6&facet.mincount=1&mm=3<-1&facet.field=schoolname_hl&facet.method=fc
> >
> > It should hopefully just take that long the first time?  How much time
> > does it take to facet on the same field subsequent times?
> >
> > > And my cache policy :
> > >
> > > <filterCache class="solr.FastLRUCache"
> > >                  size="4096"
> > >                  initialSize="4096"
> > >                  autowarmCount="4096"/>
> > >
> > >     <queryResultCache class="solr.LRUCache"
> > >                      size="5000"
> > >                      initialSize="5000"
> > >                      autowarmCount="5000"/>
> >
> > These are relatively big caches - consider reducing them if you can.
> > Especially the filter cache, depending on what percent of the entries
> > are bitsets.
> > Worst case would be 50M / 8 * 4096 = 25GB of bitsets.
> >
> > > * i'm using solr 1.4 (LUCENE_36)
> > > * 64GB Ram (with 60GB allocated to java/tomcat6)
> >
> > Reduce this if you can - it doesn't leave enough memory for the OS to
> > cache the index files and can contribute to slowness (more disk IO).
> >
> > -Yonik
> > http://lucidworks.com
> >

Reply via email to