The paper proposes a new gateway queuing algorithm, it takes the Nagle's
algorithm as the base and proposes changes on it. The main drawback in the
generally used First come first served algorithm is that it is vulnerable to
a malicious user. The user could either be the source, the destination, the
source destination pair or the individual process running on the source
host.
There is a need to distinguish the sources at the gateway routes based on
their type , a file server should naturally be given more bandwidth. But
this is not practical to implement because there is no consensus on any such
host types. Doing it on the basis of individual processes is not good
because it can be misused by end user and will also violate the layering by
letting the routers look into the port addresses.
The Nagle's round robin functions well but doesnot give fair bahavior when
packet sizes vary. To get over this problem the paper proposes a bit by bit
round robin . Since a bit by bit algorithm is computation intensive the
author proposes a packet by packet transmission scheme ( preemptive and non
preemptive). It is not clear in the paper how the non preemptive packetized
version is different from Nagel's algorithm.
The author considers three aspects of the fair queuing: bandwidth,
promptness and buffer space. The bandwidth and buffer allocation is
specified by the fairness definition proposed by the author : no user
receives an allocation more than its request. The author proposes two ways
to deal with promptness : more promptness is given to users which donot use
their bandwidth fully and faster service is given to packets that arrive at
an inactive conversation. The state of the connection for the
source/destination pair is stored to be used by the algorithm. It compares
the two types of applications ftp and telnet and shows how the delay and
bandwidth requirements are different in the two cases. FQ definitely
functions better than FCFS in this case. However its interesting that with
the introduction of new multimedia application these examples donot remain
very relevant.
Simulation:
The author combines FQ and FCFS with different flow control algorithms. It
uses a set of bench mark scenarios each of which represenst a different
congestin control aspect. The network is loaded with a set of telnet anf ftp
sources. It is clear that FQ gateways are alone not sufficient to provide
congestion control, they have to be combined with adaptive flow control
algorithms. When used with an ill bahaving source the FQ algorithm not only
ensures that every one receives its fair share of bandwidth but also that
illbehaved source is penalised for its behavior.
The paper was very interesting to read. The major criticism of FCFS in this
paper is its inability to allocate bandwidth, promptness and buffer space
independently. The conclusion is that the FQ combined with DECbit flow
control is particularly effective. There is no usage of type of service
field of IP in FQ, there are other algorithms which use this field but are
unable to allocate resources on a user-user basis, others like random
dropping buffer allocation are vulnerable to ill behaved users.