SolrCloud vs Standalone Solr

Question

I am trying to upgrade standard 4.x Solr install to Solr cloud 5.x.I did some performance testing between two and found massive differences. On the same server, at different times, I've ran the below:

• Solr cloud 5.2.1 & 5.3, 2 shards, 2 replicas, 3 zookeepers
• Solr standard Both have the same configs (solrconfig, etc).

Both have the same data

When performing an example query, multiple times, the average result was that standard Solr was 5x faster than Solr cloud.

And dataimport handler was also 2x faster than standard solr.

Any thoughts why this discrepancy and how to improve solrcloud performance?

Answer 1

We performed a similar upgrade from standalone solr to solrcloud 6.6, and we ran into similar performance issues.

What we determined is that as you add shards and replications, performance decreases, because the leader nodes have to do more work sending index data to all the shards and replications. I would not recommend using more than one shard unless your index is large enough that it can not fit on a single disk.

We ended up implementing only one shard, with multiple replications, to satisfy our use case. After tweaking the Java properties, RAM, and disk space, we were able to get solrcloud with one shard and four replications performing only slightly worse than solr standalone.

I highly recommend reading the solr documentation on solrcloud, and how it may or may not satisfy your use case. We never needed to 'upgrade' to solr cloud, so we ended up spending a lot of time on something that isn't even as good or as stable as solr standalone.