I've been working on benchmarking our solr response times in relation to the a variable number of concurrent queries. With maxThreads=150 - I've tried running between 20-100 queries concurrently against our solr instance and have noted that for all n-way (>20) queries I'm finding that performance flatlines at 20-30 requests/second.
We've tried tuning caches and while part of the poor performance is down to poor query formulation - I find the lack of seeing either performance improvement or degradation as being indicative of some kind of throttling.
Not sure if this is the case or not however as a novice in these realms I would appreciate some guidance as to what I should be looking at and where we might be able to tune/investigate?
We've ruled out disk contention and network latency.
Useful metrics: maxThreads:150 filterCache Size; 16384 queryResultCache size: 16384 documentCache size: 10502
I'm running Solr/Lucene version: Solr Specification Version: 1.3.0.2009.08.19.15.54.27Solr Implementation Version: 1.4-dev ${svnversion} - rafiq - 2009-08-19 15:54:27Lucene Specification Version: 2.9-dev
Would be grateful for any pointers and can furnish more details.