Re: [tcpm] RFC 1323: Timestamps option

You've probably already seen my reply to Sally, but I'll clarify a  
few things:

On Jan 26, 2007, at 1:42 PM, Mark Allman wrote:

>   + Why is the "best case" that you'll get multiple samples per  
> RTT?  It
>     seems to me that the underlying assumption is that "more is  
> better",
>     but I am not sure it is.  We have done some analysis that suggests
>     more samples are not better.  I don't want to say our analysis is
>     the last word here, but I'd be interested in hearing arguments for
>     taking more samples [using the current RTO estimator] per RTT  
> being
>     a clear win.
>
>     I think this is an important point in choosing a path.  I.e., if
>     multiple RTT measurements matter then that might suggest a  
> different
>     approach than if multiple RTTs buy you little/nothing.

Perhaps "best case" was the wrong term.  Being guaranteed at least  
one good RTT measurement per RTT is a good thing.  The current RTO  
estimator is based on getting one sample per RTT, no matter how much  
data flows during that time.  As you know, when you start using all  
the samples from multiple Timestamps options, if you just feed them  
into the current RTO estimator it can seriously skew the results,  
especially if you are getting lots of samples per RTT.  In my reply  
to Sally, I suggested taking the average of all the Timestamps  
received during one RTT, and feeding that into the RTO estimator.  An  
alternative would be to take the largest sample received during the  
RTT interval, and feed that into the estimator.  That'd be even  
easier to implement.  Do you have any thoughts as to which would be  
better?

Another alternative would be to only use the sample from the  
Timestamps option in the ACK for the sequence number that is being  
timed, i.e. back to one sample per RTT.  But that is still a win over  
the old method, because with Timestamps, we know that the measurement  
will always be accurate, even if there was a retransmission during  
the RTT.

But I totally agree, just feeding in the more frequent samples from  
the Timestamps into the RTO estimator is not a good thing.

		-David Borman

_______________________________________________
tcpm mailing list
tcpm@ietf.org
https://www1.ietf.org/mailman/listinfo/tcpm