Subject: Re: Interface traffic timestamps
To: Jonathan Stone <email@example.com>
From: Jason Thorpe <firstname.lastname@example.org>
Date: 06/11/2004 11:35:07
Content-Type: text/plain; charset=US-ASCII; format=flowed
On Jun 11, 2004, at 10:46 AM, Jonathan Stone wrote:
> I'd say: please don't even try to do this on gigabit or 10GbE links;
> the performance hit simply isn't worth it. (My back-of-the-finger
> estimate leaves me suspecting mono-time is too coarse to be useful.
> Or turn it around: bumping a microsecond mono_time once per input
> frame, when frames are only 50 ns long, is going to completely stuff
> up your time-keeping.
I think he's saying he wants to copy mono_time into the timestamp when
a packet is accepted. Not increment mono_time. mono_time is perfectly
sufficient for the type of application he's talking about (where the
user generally cares about "seconds", not even "microseconds").
That said, I too am concerned about the overhead of copying mono_time
on high-speed links ... to do it correctly, you actually need to do:
s = splclock()
timestamp = mono_time;
...although there are places where we skip the spl frobbing because
accuracy is not really required, just "ballpark".
> Please consider that a very loud objection. :-/. To me, that suggests
> a per-interface flag, to control whether to do the mono_time
> timestamps, or not.
So we add the overhead of testing the flag each time, too? :-) Test
and branch might be as or more expensive than just doing the memory
loads/stores (which are likely to hit the cache if done frequently
> Alternatively: is the seconds portion of (mono_)time good enough for
> your purposes? Do you really care about microseconds or milliseconds,
> for whether to drop a connect-time-charged ISDN line?
Yah, perhaps seconds is good enough ... then you pretty much definitely
do not have to worry about spl frobbing for a decent "ballpark"
-- Jason R. Thorpe <email@example.com>
content-type: application/pgp-signature; x-mac-type=70674453;
content-description: This is a digitally signed message part
content-disposition: inline; filename=PGP.sig
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (Darwin)
-----END PGP SIGNATURE-----