From: Karthik Gopalratnam (karthikg@cs.washington.edu)
Date: Sun Nov 21 2004 - 23:21:59 PST
MACAW Review
This paper proposes MACAW - an extension to the earlier MACA protocol for
establishing collsion resistant media access for wireless LANs. The authors
analyze and fix a number of problems in the MACA protocol. The first was the
the exponential backoff scheme was both unfair as well as unstable. I
thought that the authors' solution for fairness was very clever indeed -
using a "synchornized" counter for the wait time so that no one station was
able to corner all the bandwidth. The second significant contribution that
the paper makes is the ACKs on the link level. Since the medium is so lossy,
relying on the transport layer to recover from losses is clearly
inefficient, so the authors choose to rely on retransmissions on the link
layer to make the protocol more robust, which is clearly a big win in terms
of throughput and performance of the network.
While this paper addresses issues with the wireless networks that were
prevalent 10 years ago and proposes many really cool solutions, some of
which are still around today, there are some glaring holes that are brought
to the fore in the context of today's WLANs. First is the issue of
ill-behaved senders. Clearly a fault in the implementation or a malicious
sender could use the RTS-CTS packets to flood the WLAN and deny service, and
there is no in-built resilience to that kind of attack, and the authors do
not provide any discussion of this issue. THe second is that the WLAN
characteristics are much more heterogeneous today. THe 2.4-2.48 GHz
transmission frequency (3 orders of magnitude(!) more than the ones
considered by the authors) used in today's networks means that there are
significantly more multi-path propagation problems and consequently more
timing issues with the backoff algorithms - the MACAW protocol relies on
these synchronization schemes with DS packets and so forth to establish
efficient back off. Also, the clients connecting to an AP do not all send at
the same rate, but adjust their rate so taht they compensate for their
signal strength, which can vary widely. Therefore, those clients that send
at low data rates end up taking more air time, which makes it unfair for
faster clients. (This is something that we believe we have an elegant
solution for in the general case, as part of our project, where we can
consider multiple classes of service and be fair to these classes in terms
of air time as well).
This archive was generated by hypermail 2.1.6 : Sun Nov 21 2004 - 23:22:00 PST