From: Tom Christiansen (tomchr@ee.washington.edu)
Date: Sun Nov 21 2004 - 22:46:01 PST
In this paper a protocol for communication over wireless networks is
designed. The protocol is largely based on the MACA protocol Phil Karn
designed in 1990 for use on packet radio. However, as pointed out in the
article, the MACA protocol has several shortcomings when it comes to
collision avoidance and congestion control. The MACAW protocol solves many
of these issues. One problem with the MACA protocol is the back-off timer
which is doubled after every collision (collision = RTS is sent but no CTS
is received). This scheme results in an unfair allocation of bandwidth
between users on the network as one transmitter will always maintain a
relatively low BO counter whereas the remaining users will have BO counters
maxed out at the upper limit. These unfortunate users will only be able to
transmit a request to send during the small pause between the time one data
packet ends and the time the "bandwidth hogging user" sends the next RTS.
This problem is addressed by copying the BO counter between different nodes
on the network, thus, resulting in a fair bandwidth sharing.
Several other issues are presented, but the common pattern is that the
issue arises because the different nodes on the network cannot hear each
other and, therefore, can not synchronize or schedule data transfer. The
proposed solutions have in common that they relay the current network state
(or transmission state) to the users to allow the users to take their
precautions.
The developed protocol is evaluated through simulations. MACAW is
significantly better than MACA over noisy communications channels. It also
provides better bandwidth sharing. However, in an ideal environment (no
noise sources), MACAW provides slightly lower throughput. This is probably
due to the larger overhead and handshaking algorithms of MACAW.
The congestion control algorithm still balances on one point: A collision
is detected as a missing CTS (transmitter sends RTS and waits for CTS
before sending the data). However, there are many reasons why a CTS could
be missing. This issue is briefly mentioned in the article but not covered
in any great depth. I wonder if it was possible to design a better
collision detection scheme using the DS packets.
The article focuses on what to do when things go wrong. But makes no
mention of what happens when things go right (or sort of right). On HAM
packet radio one problem is (or was - I honestly don't know if it is used
anymore) that the users were expected to maintain a "quiet period" on the
radio before transmitting. This presents two issues: 1) Unless the timer is
randomized (like it is in ethernet) all users will transmit at the same
time resulting in collision. 2) On packet radio it was quite common that
"aggressive"/misbehaving users would set their "quiet period" timer rather
low in order to be the first to transmit. This issue is also left out in
the article.
Overall it is a well-written article which is interesting to read. One
could argue that it only presents an incremental change to the MACA
protocol, but in that case the increment would be rather large after all.
The protocol design is backed up by data showing good performance.
This archive was generated by hypermail 2.1.6 : Sun Nov 21 2004 - 22:46:17 PST