Re: Simultanous PID!?

["Thomas 'Dent' Mirlacher" <dent@cosy.sbg.ac.at>]

On Tue, 30 Apr 2002, Kearney wrote:

> what about the processing overhead?
> How many machine instructions and memory references would it take on a
> "typical" modern processor to decide whether this is a TSpacket which should
> be deliverd to the next upper layer or simply dropped - i.e. do the saem
> thing a hardware interface does?

well the calculations are pretty basic. - if someone's interrested we could
come up with a pseudo-fragment to filter the data.
- in any case: this kind of filtering compared to the other layers within
	the network-stack of a modern operating system is for free (compared
	to ops used in other layers)
- it's a different story for building an embedded system ... (router/switch)

here some simple figures:

at 100Mb the filtering has to be done 70000 times a second
(for currecntly common transponders this would be 1/3rd ~ 23000)
... and this multiples by the number of ops needed. (granted the
4004 wouldn't be enough :)

but on the other hand: (which is kind of off-topic) ... wouldn't
	you consider it useful to have the SAR and filter on the
	card/in the FW. this would decrease this "background-noise"
	plus you gain something by not having the TS-overhead to
	transport via the PCI/whatever bus.

	just my $0.02

		++Thomas
-- 
in some way i do, and in some way i don't.