From: Daniel Lowd (lowd@cs.washington.edu)
Date: Wed Oct 27 2004 - 02:31:45 PDT
This paper discussed ways to offer two different types of real-time
application service, together. One of the most important contributions,
however, is simply a discussion of the types of service needed and that
different queueing algorithms serve each best. The observation that FIFO
gives better service than WFQ for adaptive applications was not something
I had considered, but it makes sense. The authors also discussed FIFO+
and ways of combining them: simply run FIFO+ as one WFQ flow.
While providing an important foundation for future work, this paper
suffered from a number of flaws that may limit its short-term impact.
First, it wasn't clear where this technology would be deployed, or what
the deployment strategy would be. It seems unlikely that the Internet,
with its distributed management, would be able to coordinate the types of
guarantees and enforcement that the proposed schemes require.
Furthermore, if enforcement was not strict, then what would stop a packet
from lying about its priority in its header?
I was also frustrated by the handwaving description of the protocol and
the limited experimental verification. (It seems that doing a good
experimental evaluation is difficult in general, since networks can vary
so much.) Perhaps I would have minded all of this less if the article
itself had been shorter.
Even so, this paper offers a good set of definitions and initial
perspective from which future work can build practical protocols for
future real-time applications.
-- Daniel
This archive was generated by hypermail 2.1.6 : Wed Oct 27 2004 - 02:31:45 PDT