Henning Brauer wrote: > * Anton Maksimenkov <[email protected]> [2010-10-13 11:33]: > > BTW, in worst case (I used ping -f, just can't find anything useful to > > generate many small packets) "re" generate about 5000 interrupts. > > While "vr" on same machine can do about 20 000 interrupts and the > > machine still stay. > > the interrupt counts say nothing really. you don't know if there is > any interrupt mitigation in action, and if so, they do differ between > the various chips, too.
Am I understood correctly that mitigation occurs more frequently under high load and has no accountable effect on moderate load (i.e. more load, more interrupts get mitigated and vice versa)? If so, is 5k interrupts that high to take mitigation in account? 20k? Besides of that, some cards have TX-only coalescing which means no interrupt usage optimization if your RX pipe is overloaded. In addition, earlier Realteks used fixed size ring buffers (not sure is it still true for newer re(4) chips and directly related to the problem). Alexey
