I need to implement a server to which multiple clients can send requests simultaneously. The code processing an individual request might block (with the thread going to sleep) in the middle.
At the moment I am using the C++ GRPC synchronised server. Each time a client sends a request, a new thread is spawned on the server's side. This is a problem since a server can create too many threads simultaneously.
I am considering two solutions to avoid the problem:
1) Use the sync server with the ResourceQuota (e.g. restrict the max number of threads to 10).
2) Use the async server.
Implementing the second solution is considerably more difficult than implementing the first solution. What advantage (if any) the second solution would give compared to the first one? Which solution would give better results in terms of:
- The amount of time an individual client needs to wait to get a response to RPC
- The resources (memory, threads) used on the server.