I do want to point out that interrupt coalescing also has its draw backs. Many people discussed all the good things so lets underscore the bad. Networks that have trickling amounts of packets are hurt by coalescing. Most notably are interactive things like typing in ftp and ssh. Since the card is going to collect several packets you basically always hit the coalescing timer slowing the overall throughput down to whatever your timeout is.
The big issue with coalescing is that it is good for some loads and bad for others. So if you can't predict your load it will either hurt you or help you. So far I have never seen a heuristic that works well both ways. Also what I have seen is that coalescing always hurts disk IO performance. Every time I tinkered with this it had a negative impact. The caveat here is that NICs generate interrupts at a completely different order of magnitude. On Fri, Sep 07, 2007 at 02:31:15AM +0000, Sam Fourman Jr. wrote: > hello misc@ > from the page http://www.openbsd.org/42.html , one of the changes made > to OpenBSD 4.2 is > > A change in the way the kernel random pool is stirred greatly > increases performance with network interface cards that support > interrupt mitigation, especially on architectures where reading the > clock is expensive (such as amd64). > > What would be some Examples of Network Cards that Support "interrupt > mitigation" > > I guess on this Subject I need educated because I am not all together > sure what interrupt mitigation is and why I want it. > > > Thank you for another GREAT release > > Sam Fourman Jr.
