From: Tom Christiansen (tomchr@ee.washington.edu)
Date: Tue Oct 26 2004 - 18:49:51 PDT
This article deals with the motivation for guaranteed quality of service,
characterizes the main challenges with QoS (delay, jitter), and proposes a
new scheduling algorithm, FIFO+. This algorithm is compared against
existing algorithms and simulated under different load scenarios.
The article gives a good background analysis of the reasons for
establishing network services with different priorities. It gives good
examples of where real-time may be applicable. The implemented FIFO+
algorithm provides a 40-50 % improvement in delay compared to other
scheduling algorithms (FIFO, and Weighted Fair Queuing).
The article, while being reasonably well-written, is incredibly long and
wordy. Unfortunately, the FIFO+ algorithm isn't described in any level of
detail which would enable someone to implement it an replicate the results
of the article. This is highly unfortunate.
The article, conveniently, avoids dealing with the issues of admission
control. It briefly mentions that some control scheme should be used, but
that's about it. One method for selecting different levels of service is
already in place in form of the TOS field in the TCP header. But there is
nothing which prevents an ill-behaved source from setting a high-priority
bit and getting premium service. In reality, user authentication is
probably the main challenge of QoS - at least with current protocols.
This archive was generated by hypermail 2.1.6 : Tue Oct 26 2004 - 18:49:59 PDT