lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Nicolas Bélisle <nicolas.beli...@doculibre.com>
Subject Performance questions
Date Tue, 28 Nov 2017 19:50:50 GMT
Hi,


We use Solr like a search engine / document store / database. We are
currently optimizing a test environment and would welcome any relevant
suggestions.

I've taken a lot of time researching this mailing list and found a lot of
relevant information.


Here's our current setup :


SolrCloud 6.5 Cluster : 5-6 nodes (8 cpu, 16GB ram : 8GB Solr and 8GB OS)
with 3-4 Solr cores each. Shards: 9-10 with replicationFactor = 2.

Current sharding: we have 4-5 cores per 8-cpu server.

Documents : 10M +, based on Wikipedia.

We use dynamic fields, with hundred of different fields (400?) but
individual documents have around 50 fields. We store most field.

We do not commit immediately: once every few seconds

We use fast hybrid storage (1 GB/s write)

Cache : generally have 0.4-0.66 hit rate. Default options.

Tried changing max indexing threads and merge factor. No significant gain.

We use an application cache for simple queries (get by id)

Current performance: on a 1100 concurrent users load, we average 3-4
seconds by query / update.

We use a couple of negative filter queries. Example *:* AND -type_id:A

Couple of questions:

-          Suggestions for new optimisations?

-          Anyone seen performance gains from 6.5 to Solr 7.1?

-          Do negative filter queries severely impact performance?

-          Heap is not an issue, but could docValues enhance search
performance? Or just memory usage?

Here’s an excellent guide we used for some of our work :
http://events.linuxfoundation.org/sites/events/files/slides/HighPerformanceSolr.pdf


Nicolas

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message