Hi Noticed that a solrcloud cluster doesn't scale linearly with # of nodes unlike the unsharded solr cluster. We are seeing a 10 fold drop in QPS in multi sharded mode. While some overhead is understandable, the pattern that we see is, some nodes load averages shoot up and that in turn skews up the overall latencies of the cluster since those nodes become the weaklings in the system.
We tried balancing things out by having atleast a shard(collection) for each node but there is still lot of unevenness in load averages across shards. Is there a way to avoid these kind of hotspots? Suchi