Re: [patch] softirq-2.4.10-B2

[Ingo Molnar]

On Fri, 28 Sep 2001, Simon Kirby wrote:

> Actually, I just tested my while(1) sendto() UDP spamming program on
> an Acenic card and noticed that it seems to do some sort of batching /
> grouping, because the number of interrupts reported by vmstat is only
> 8,000 while it's easily 75,000 on other eepro100 boxes.
> Interesting...

yes, acenic.c (and the card itself) supports irq-rate limitation and
work-coalescing. Almost all gigabit cards do. Check out the tx_coal_tick,
rx_coal_tick, max_tx_dsc, max_rx_desc tunables that can be set at insmod
time. Note that this does not decrease the amount of work generated, but
it will reduce the irq-processing overhead significantly.

The eepro100 card does not provide such ways - there the only way to stop
future interrupts from happening is to stop the receiver, or to not refill
the rx-skb queue. (or disable the interrupt, which is a pretty crude and
inaccurate method.)

	Ingo