What do people think about adaptive polling?

Matthew Dillon dillon at apollo.backplane.com
Tue Dec 12 11:04:51 PST 2006


    Splitting the work into two halves really only applies to a network
    interface interrupt, and we actually already do it because our 
    protocol stacks are threaded.  There are clear benefits under heavy
    network loads because our methodology seems to make livelocks impossible
    and seems to result in far smoother transitions when going from a loaded
    to an overloaded state.

					-Matt





More information about the Kernel mailing list