[e2e] pre-BOF discussion: "ABR to the Internet"

Michael Welzl michael at tk.uni-linz.ac.at
Thu Jun 28 14:35:07 PDT 2001


> Although today's CPUs are capable of indicating elapse time 
> at the CPU clock
> resolution using a single instruction, the OS and networking interface
> depends on much lower time resolutions.  As such, setting 

All I know is that there is an "ifOutOctets" available, which is probably
implemented in hardware in my network interface card. Is this resolution
really a problem, given that the information can only reach the end nodes
every RTT?

In any case, it can't be much worse than trying to implement a TCP-friendly
mechanism on top of UDP under Windows NT (as we did).   :)
We ran into all kinds of trouble with timer granularity.

> rates versus the
> self pacing scheme will demand that a high resolution of time be made
> available and that transmissions can be scheduled based on 
> rate intervals.
> This would seem to be a major departure from the present techniques.
> Available Bit Rate (ABR) could translate into a send cap of 
> some type, but
> it will be difficult to ensure any actual use without a higher time
> resolution than typically available to the drivers in my opinion.
> Doug

Well... I can't really say much about that  :(


More information about the end2end-interest mailing list