Hello All, I've been working on trying to send multiple requests in parallel from an NFS client over UDP to a server. We're working on improving the performance of NFS on our little core with lots of threads. Given this environment, my goal is to send rpc tasks over UDP from more than one thread at the same time. To do this, I take advantage of some of the code changes made in the TX path that allow the sending of RPC packets without having to cork the socket when MSG_MORE is false. With this change, I don't have to lock the transport allowing me to send rpc_tasks from multiple threads at the same time (there is some serialization in the driver down below, but that seems okay). Without this change (requests are sent out serially one after another) the server gets a request, formulates a reply (my file is probably cached on the server) and sends out a response in anywhere between 40 and 90 microseconds. It then sees the next request and so on. With this change, the client is able to "flood" the server with a lot of requests very quickly (they arrive about 10-15 microseconds apart). The responses to the clients request get delayed and in some cases the response gets delayed by 800 microseconds. Is there a way to increase the number of threads that service a single client connection? Thanks -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
