Thanks for the reply. The overhead you describe is what I suspected, I was just suprised that if DSE is able to keep that overhead small enough that the overall result is faster with the extra hardware, Solr doesn't also benefit.
I did try with RF=2 and shards=1, and yep, it's way fast. Really nice performance. I'm hoping I can figure out a configuration that will allow me to get this type of performance boost over DSE with a much higher load. Follow up question: I'm getting periodic errors when adding documents in the Java client app: org.apache.solr.client.solrj.SolrServerException: IOException occured when talking to server at: http://<ip address>:8983/solr/<collection name> I believe it's always preceded by this log message: Request to collection <collection name> failed due to (0) org.apache.http.NoHttpResponseException: <ip address>:8983 failed to respond, retry? 0 I'm guessing it's due to a timeout, but maybe not my client's timeout setting, but instead a timeout between the two Solr Cloud servers, perhaps when a query is being sent from one to the other. Probably unrelated to those errors because these are at different times, but in the Solr logs, I'm getting occasional warning+error messages like this: always server 2: WARN [c:<collection name> s:shard1 r:core_node2 x:<collection name>_shard1_replica1] o.a.s.c.RecoveryStrategy Stopping recovery for core=[<collection name>_shard1_replica1] coreNodeName=[core_node2] followed by server 1: ERROR [c:<collection name> s:shard1 r:core_node1 x:<collection name>_shard1_replica2] o.a.s.u.StreamingSolrClients error org.apache.http.NoHttpResponseException: <server 2 IP>:8983 failed to respond <repeated 14 times, so 15 total> WARN [c:<collection name> s:shard1 r:core_node1 x:<collection name>_shard1_replica2] o.a.s.c.LeaderInitiatedRecoveryThread Leader is publishing core=<collection name>_shard1_replica1 coreNodeName =core_node2 state=down on behalf of un-reachable replica http://<server 2 IP>:8983/solr/<collection name>_shard1_replica1/ Any idea what the cause is for these and how to avoid them? -- View this message in context: http://lucene.472066.n3.nabble.com/Solr-Cloud-Higher-search-latency-with-two-nodes-vs-one-node-tp4295894p4296124.html Sent from the Solr - User mailing list archive at Nabble.com.
