From: Craig M Prince (cmprince@cs.washington.edu)
Date: Wed Oct 27 2004 - 05:21:10 PDT
Reading Review 10-27-2004
-------------------------
Craig Prince
The paper titled "Supporting Real-Time Applications in an Integrated
Service Packet Network: Architecture and Mechanism" provides a means of
supplying various levels of Quality of Service in a network. The paper
begins with an analysis of the types of realtime traffic that their
protocol needs to support. They then go on to describe two separate
protocols, one that provides the traditional guaranteed service and one
that provides what they call predictive service. These protocols are then
finally combined to produce a single protocol supporting three types of
service (the two above, plus traditional best-effort service). This
component of the QoS architecture was the focus of this paper with very
little attention given to how sources specify their flow and how sources
are admitted to the network -- two fundamental issues in any quality of
service implementation.
The neat part about this paper is that it correctly identifies that not
all of the real-time traffic needs the strict bandwidth/delay guarantees
and so provides various levels of service directly within the same
protocol. The highest level does of course provide the strict
bandwidth/delay guarantees; however, there is a second level that provides
what is called predictive service that tries on average to provide a
certain level of bandwidth and delay. This is adequate for applications
that can tolerate some loss and some changes in delay (which is the bulk
of most real-time applications).
The analysis of the FIFO+ protocol, while not rigorous, was adequately
convincing in its arguments -- the cool observation being that FIFO in
general does a good job of sharing the jitter amongst various sources,
which is what we went if we want to reduce individual jitter.
Unfortunately, this property only holds for well-behaved sources. And I
don't believe this work adequately addresses the issue of isolation for
these sources (although there is mention of checking this at the edges of
the network).
I liked the fact that this article touched on the economic issues related
with Quality of Service; however, I this paper did not convince that their
architecture could be successfully deployed on the internet today. The
heterogeneity and economic factors associated with the internet make such
deployment nearly impossible. There has been little work done on creating
a QoS system that is actually mutually beneficial to ISPs
Another issue I had with this paper is that it did not fully explore the
issue of admission control (another barrier to actual deployment). While
not the focus of this work, admission control is vital to a successful QoS
scheme yet is often overlooked.
Overall, this paper provides a wealth of possible research directions. The
biggest is the creation of QoS systems that provide economic incentives to
ISPs so as to hasten their adoption. This paper provides a mechanism for
delivering QoS, but does not have the required accounting, etc. to make it
viable. However, this paper shows that it is feasible to build an
architecture with the varying levels of service needed.
This archive was generated by hypermail 2.1.6 : Wed Oct 27 2004 - 05:21:11 PDT