I was exploring the replication options in SOLRCloud/ES and thought would
share my findings here
1. There is a notion of master/slave(s) for every shard
2. Every operation is separately computed on master+slave(s)
3. Failover/Sync-Code is extremely hairy in both SOLR/ES...
In blur, only the server responsible for shard does the computation. There
is a huge reduction in resource consumption cluster-wide {I/O, CPU, Memory
etc...} when compared to SOLR/ES etc... Failover/Sync is also quite
clear/easy to understand...
For a write-heavy system like ours, this is a huge boost...
If my observation is correct, can we include this in blur-wiki?
--
Ravi
