On Mon, May 26, 2008 at 2:56 PM, Pierre Lamy  wrote:

quoted text
> Claudio Jeker wrote:

>> Polling is a hack that works only nicely around a particular load. It

>> fails for high and low loads. Using good interfaces with correct interrupt

>> mitiagtion on a fast bus is way better then polling.

>> Polling sounds very nice on the paper but only there. For a system

>> handling more then a few thousand packets per second you need to increase

>> HZ in such rediculous hights that it will break something for sure.

>> FreeBSD users suggest to set HZ to 10000 and more on system with higher

>> network loads. IMO that's insane.

> 10k hz is 10,000 cycles per second. What's that, like 10 milliseconds?
since when 1 / 10000 s (i.e. 1e-4 s) is 10 miliseconds (i.e. 10e-3 = 1e-2)?

10 kHz is 100 microseconds (100e-6). now
> The numbers above may be off by several orders of magnitude, I've been up 40

quoted text
> hours.

uh yeah, what are those calculations then?
> Pierre