Subject: Re: Interface traffic timestamps
To: Martin Husemann <martin@duskware.de>
From: Steven M. Bellovin <smb@research.att.com>
List: tech-kern
Date: 06/11/2004 18:55:32
In message <20040611224826.GB16781@drowsy.duskware.de>, Martin Husemann writes:
>On Fri, Jun 11, 2004 at 03:42:43PM -0700, Jonathan Stone wrote:
>> So wrap it in a macro. Write a macro (or inline function in sys/net/if.h)
>> which takes a struct ifnet *, and DTRT. Whatever that may become, over time
>
>Another idea maybe worth to think about: actually this is only interesting on
>links where filtering is active (i.e. some packets could be dropped by pfil
>and whatever sits behind that). Maybe doing some timestamp related operation
>only if pfil dropped a packet could avoid the global performance hit.
>
I think that that would violate the rule of least surprise. Perhaps
something else will look at that field some day, and it won't work
right. Jonathan is right about the cost of fine-grained timers here,
but for the coarse timer we're talking about it should be extremely
cheap to do what you originally asked for.
--Steve Bellovin, http://www.research.att.com/~smb