3 ms·
The standard rule for CPU-bound RPC server utilization is 80%. Any less and you could use fewer machines; any more and latency starts to take a hit. This is whe
by dehrmann 2y ago
The standard rule for CPU-bound RPC server utilization is 80%. Any less and you could use fewer machines; any more and latency starts to take a hit. This is when you're optimizing for latency. Throughput is different.
- pdhborges 2y agoDoesn't this depend on the number of servers, crash rates and recovery times? I wouldn't feel confident running 3 servers running at 80% capacity in ultra low latency scenarios. A single crash would overwhelm the other 2 servers in no time.
- dehrmann 2y agoRight; this is only for large pools of servers.