----- Original Message ----- From: "Oliver Kowalke" <k-oli@gmx.de> To: <boost@lists.boost.org> Sent: Wednesday, November 05, 2008 10:25 AM Subject: Re: [boost] [threadpool] mixture of recursive invokation and fibers?
Imagine we have stage A which reads message from the network and stage B processing the messages read by A. If both use the same threadpool and mltiple clients send a large messages (ready will take some significant time) stage A coul possibly allocate all threads in the pool so not work-item in stage B can be executed. If both stages use different pools the processing of work-items is independent. B can executes its task even if stage A becomes a high load.
Why the A would allocate all the the threads in the pool? If the stage A handle a message and the request B to do somthing the task of B will be enqueued before other new messages comming from the network, so B will be interleaved with A. What I'm missing?
data-flow: network -> stage A -> stage B -> network If 10.000 clients are connected to the service and 1% of the clients send a large message then stage A is triggered to read/process 100 requests at the same time - possibly all threads of the pool are used by stage A and tasks of stage B are queued in the pool until a worker-thread are finished with items of A. ___________ Maybe, I would like to see some benchmarks. Vicente