I am doing some load tests locally with and without request cache turned on for a single instance of ES. I don't see any improvements with the qps and latencies we can handle with the cache turned on. I've confirmed with the _stats/request_cache api that even with 100% hits and 0% misses (after cache is warmed), there are no apparent differences. CPU and memory and latencies are all still poor. I know my application is CPU bounded due to our expensive queries, and I've optimized them as much as I can. Shouldn't the request cache help reduce CPU load?
It feels like something is wrong. Is there anything I should look into?
Looks like everything is working for the cache:
"primaries" : {
"request_cache" : {
"memory_size_in_bytes" : 1510590,
"evictions" : 0,
"hit_count" : 30322, <--- keeps increasing
"miss_count" : 2460 <--- stationary
}
},