From: Alan L. Liu (aliu@cs.washington.edu)
Date: Sun Nov 21 2004 - 21:41:45 PST
The paper presents incremental changes to a media access protocol called
MACA for wireless LANs that resolves some issues of fairness and
bandwidth utilization.
The paper identifies how traditional protocols are not suitable for
WLANs because of wireless-specific problems. For example, the hidden
terminal problem is caused by the fact that a node can only hear a
subset of other nodes, although nodes it can't hear can also communicate
with common base stations and therefore cause trouble. The exposed
terminal problem is caused by simultaneous use of the spectrum being
possible if two nodes talk to two different base stations. Carrier sense
is not enough to overcome either of these. Therefore, a WLAN protocol
must distribute the burden among both the senders and the receivers.
Some clever tricks with sending control packets to avoid unfairness are
described. The DS message enables those nodes who didn't hear the entire
RTS-CTS exchange to know that the channel will be used, and also for how
long. Without this, a node must be lucky in order to acquire the
network, because its backoff time must fall right in the small time gap
between long data transmissions.
One interesting thing they introduced in WLAN was the use of ACKs for
reliable transmissions, below the TCP layer. Apparently this is because
recovery is so much faster from the link-layer. When the error rate of
the medium is somewhat high, this achieves far greater throughput than
without ACKs.
One thing I found lacking about the paper's analysis of MACAW is that
the paper gives no "typical" WLAN network attributes. It is hard to know
whether the performance gain of using MACAW at 1% packet loss rate is
worth the performance loss at 0.1%. I would have preferred that the
paper gave some insights into the wireless characteristics before
attempting to simulate such networks. Even better, real data could have
been used. I understand that it is prohibitive to have actual
implementations of link-layer protocols because of the need for part of
it to go into hardware. This seems like it could be addressed by efforts
like STP, so that hardware doesn't have to be the bottleneck for getting
real data.
One gaping hole in MACAW is that fairness is only ensured if all
clients cooperate and have the correct implementation. The authors did
not even consider rogue nodes. Although total prevention of a malicious
person jamming the airwaves might be impossible, I feel that some
consideration should have been made at least to robustness in the face
of badly made implementations.
It's funny how appropriate it was to write my Cerf and Kahn TCP/IP paper
review and send it over TCP/IP to the mailserver. In the same vein, I
happen to be writing this on my laptop which is "connected" using 802.11
to the Internet. It works alright, but there are definitely some
mysterious aspects of its performance that until reading this paper I
only vaguely understood. The paper allowed me to appreciate some of the
subtleties regarding wireless performance and sharing. Despite being
more complicated than Ethernet, it has recently seen an exciting growth
in bandwidth capability. Wireless usage is definitely here to stay and
prosper.
This archive was generated by hypermail 2.1.6 : Sun Nov 21 2004 - 21:41:55 PST