Without seeing the queries, it's hard to say. There was a faceting issue with some Solr versions that you might be hitting. The first thing I'd try is just straight-up 5.4, non cloud, non-sharede to compare against 4.7 to see whether it's just a difference between 4.7 and 5.4.anything to do with it, then tune if so. _Then_ go to SolrCloud.
Sharding should _only_ be used when there are too many documents to fit into a single shard and give you good results. Sharding inevitably adds overhead to the query time (although nothing like what you're seeing so that's a different problem). You can easily use SolrCloud with a singly-sharded collection with replication and get all the goodness of failover, HA/DR etc. after you determine whether your perf issues are related to differences (perhaps you've tuned the 4.7 setup differently?). Best, Erick On Fri, May 6, 2016 at 12:44 AM, Bismaya Vikash <biswaya_vi...@apple.com> wrote: > Hi everyone, > > We are trying to migrate from a > > standalone Solr 4.7.0 with 1 collection(no sharing no replication) and 3GB > of heap memory > to a > SolrCloud configuration with a 3-Zookeeper ensemble, 2 Solr instances with 1 > collection(2 shards per collection and a replication factor of 2). > We have the two Solr instances on two different servers A and B, with 1 > zookeeper on server A and other 2 on server B. > We have assigned 4GB heap memory to each Solr instance. > > > Below are the config files I am using. > > > > Zookeeper Config: > tickTime=2000 > initLimit=10 > syncLimit=5 > > We have less than 400,000 documents indexed and we are using > a singleton CloudSolrClient instance from Solrj 5.4.1. to index/search over > the collection. > > However, our search/indexing performance is slower than the standalone > version. > > Standalone Solr SolrCloud > Time taken: avg. 1000ms varies from 4000 to 15000 ms. > > Following are our queries: > > 1. Is SolrCloud slower compared to standalone Solr ? > 2. How can we improve the search time to less than 1000 ms ? > 3. Should we use a singleton CloudSolrClient or create a new instance for > every new request ? > > Thanks and Regards, > Bismaya >
