Currently these are some numbers of very lightweight advance requests (without forks) made through local/jsonrpc/grpc machines:
local 2819.05 req/s
jsonrpc 491.62 req/s
grpc 69.71 req/s
Local is way faster than jsonrpc, jsonrpc adds the overhead of:
Networking
Establishing a new connection for every machine method call
JSON serialization/deserialization
Possible solutions
We can optimize jsonrpc requests during a remote machine session by using keep alive connections, if that is not enough, we should also investigate if there are simple ways to optimize the serialization or networking.
Context
Currently these are some numbers of very lightweight advance requests (without forks) made through local/jsonrpc/grpc machines:
Local is way faster than jsonrpc, jsonrpc adds the overhead of:
Possible solutions
We can optimize jsonrpc requests during a remote machine session by using keep alive connections, if that is not enough, we should also investigate if there are simple ways to optimize the serialization or networking.