Programming

Hi,

I have a program which needs to connect multiple servers at the same time. The program has to collect data from each of servers and then make a decision regarding to the data received.

There are several requirements.
1. Server (s) may shutdown anytime without any ack (e.g.power off).
2. Client has to connect several servers.
3. Every decision should "discuss" with servers; if one of them failed, the decision will based on others which still alive.
4. The discussion should be fast, more than 200ms is not acceptable.

My design principle is like that:
1. Create threads which every one connect to a server.
2. Creating a sending queue on each thread.
3. If the queue doesn't empty, then connect the server and send the data from queue to server.
4. Wait message from server.
5. Close the connection

There are also problems in the design.
1. If I sent the message to server, how about if the server dead and no response? (receive timeout?)
2. If we create and initiating socket in a short time, a lot of TIME_WAIT since the client will close the socket when send and receive operation finished.
3. Since I have to connect multiple servers, if one of them are failed. I have to reconnect them, may say every 1 second a try?

Would you please give me some ideas that how to handle such kind of application? Thanks a lot!

TCP timeouts can take entire minutes. Best-effort delivery and all that.

I'd try a simpler system using UDP. The client sends a 'hello' to a bunch of servers via one packet each then does recvmsg() in a loop until 200 milliseconds are up. Any servers that answered in time get considered.

As an aside, that 200 milliseconds requirement isn't quite sane. Many people and services use connections that have a higher minimum latency than that. Anything that's actually busy and not idle could add latency too.