[e2e] Simple Question on TCP Window Size
Eric A. Hall
ehall at ehsco.com
Wed May 9 21:24:14 PDT 2001
> My point about overselling was intended to be about DNS and other very
> low volume (per given pair of hosts) UDP/IP applications.
My point was to refute the example cited. To say that DNS is a shining
example of what not to use with congestion feedback is not necessarily
accurate, since it can be useful for DNS implementations that already use
other feedback mechanisms for server selection, and since DNS is designed
to use feedback for server selection. Obviously there are developers who
think measurements are useful, and the ones who want to maximize their
input controls (and who already do latency sampling and ICMP error
monitoring, among others) could certainly use this as an additional
weighting value. That's not selling, it's making a point.
As to your reasoning that it is unsuitable because the metric rarely
changes, to me that is an argument against other mechanisms which are
already commonly used, such as decay timers which force periodic latency
re-evaluations. For those systems that use these tools, the congestion
signals can either supplement or short-circuit the timer. For those
systems that don't, certainly ECN would be of no interest, and would be
ignored along with all of the other stuff (yes there are DNS servers that
even ignore ICMP errors).
You may not find it useful, and there are likely to be plenty of people
who would agree with you. But to say that this scenario is overselling the
features because you don't think it's useful is projecting at best, and a
mischaracterization at worst. It can be useful to the servers that find
these kinds of inputs useful. DNS in particular is well suited to this
because of its design. Point.
Eric A. Hall http://www.ehsco.com/
Internet Core Protocols http://www.oreilly.com/catalog/coreprot/
More information about the end2end-interest