otis_gospodnetic at yahoo
Dec 16, 2011, 9:17 AM
Post #2 of 2
Re: Performance issues with high query volume
[In reply to]
It's hard to tell. For example, the problem could "simply" be with the JVM and how it is tuned or not tuned, not at all a Solr problem. Or it could be a Solr problem.
Try using SPM (link below, it's free) - this will help you figure out what is going on and see what effect any changes you make to Solr have.
Performance Monitoring SaaS for Solr - http://sematext.com/spm/solr-performance-monitoring/index.html
----- Original Message -----
> From: rstaudinger <rstaudinger [at] gmail>
> To: general [at] lucene
> Sent: Monday, December 12, 2011 6:00 PM
> Subject: Performance issues with high query volume
> I inherited a fairly large SOLR installation that has some performance
> issues in terms of query times. Since I am newbie to SOLR, I am looking for
> pointers where to start looking for the problem. Here is the basic setup of
> the system:
> - sharded across 4 servers (24 CPU, 48GB RAM)
> - the index size on each shard is 11GB
> - each shard contains about 4.5 million documents
> The system receives a pretty constant load of search requests, at a rate of
> about 110 requests/second. I can control the rate of request sent to SOLR,
> and this is about the best rate I can get, which CPU maxed out on all 4
> servers. Average search time is around 4000ms on the master and anywhere
> from 700ms to 1500ms on the slaves. Search times drop proportionally with
> the number of requests. The best search time average I can get is around
> I suspect that this configuration is not appropriate, since the sharding is
> geared more toward large datasets, not so much query volume, and a
> master-slave configuration might be more appropriate. Is an index size of
> 11GB appropriate, or should it be larger or smaller in this situation?
> We are also on SORL 1.4.1, which might be part of the issue as well. I am
> planning to upgrade to the latest version in a few months. Are there
> performance improvements from that alone even with the current
> Any pointers to either narrow down the problem or suggest alternate
> configurations would be greatly appreciated.
> View this message in context:
> Sent from the Lucene - General mailing list archive at Nabble.com.