This preview shows page 1. Sign up to view the full content.
Unformatted text preview: in popularity, uses radio. Examples include cellular wireless networks (including standards like EDGE, 3G, and 4G), wireless
LANs (such as 802.11, the WiFi standard), and various other forms of radio-based communication. Broadcast is an inherent property of radio communication, especially with
so-called omni-directional antennas, which radiate energy in all (or many) different directions. However, radio broadcast isn’t perfect because of interference and the presence of
obstacles on certain paths, so different nodes may correctly receive different parts of any
given transmission. This reception is probabilistic and the underlying random processes
that generate bit errors are hard to model.
Shared bus networks. An example of a wired shared medium is Ethernet, which when
it was ﬁrst developed (and for many years after) used a shared cable to which multiple
nodes could be connected. Any packet sent over the Ethernet could be heard by all stations
connected physically to the network, forming a perfect shared broadcast medium. If two
or more nodes sent packets that overlapped in time, both packets ended up being garbled
and received in error.
Over-the-air radio and television. Even before data communication, many countries in
the world had (and of course still have) radio and television, broadcast stations. Here, a
relatively small number of transmitters share a frequency range to deliver radio or television content. Because each station was assumed to be active most of the time, the natural
approach to sharing is to divide up the frequency range into smaller sub-ranges and allocate each sub-range to a station (frequency division multiplexing).
Given the practical signiﬁcance of these examples, and the sea change in network access
brought about by wireless technologies, developing methods to share a common medium
is an important problem. LECTURE 10. SHARING A COMMON MEDIUM: 4 MEDIA ACCESS PROTOCOLS 10.2 Performance Goals An important goal is to provide high throughput, i.e., to deliver packets successfully at as
high a rate as possible, as measured in bits per second. A measure of throughput that is
independent of the rate of the channel is the utilization, which is deﬁned as follows:
Deﬁnition. The utilization that a protocol achieves is deﬁned as the ratio of the total
throughput to the maximum data rate of the channel.
For example, if there are 4 nodes sharing a channel whose maximum bit rate is 10
Megabits/s,2 and they get throughputs of 1, 2, 2, and 3 Megabits/s, then the utilization
is (1 + 2 + 2 + 3)/10 = 0.8. Obviously, the utilization is always between 0 and 1. Note that
the utilization may be smaller than 1 either because the nodes have enough offered load
and the protocol is inefﬁcient, or because there isn’t enough offered load. By offered load,
we mean the load presented to the network by a node, or the aggregate load presented to
the network by all the nodes. It is measured in bits per second as well.
But utilization alone isn’t sufﬁcient: we need to worry about fairness as well. If we
weren’t concerned about fairness, the problem would be quite easy because we could arrange for a particular backlogged node to always send data. If all nodes have enough load
to offer to the network, this approach would get high utilization. But it isn’t too useful in
practice because it would also starve one or more other nodes.
A number of notions of fairness have been developed in the literature, and it’s a topic
that continues to generate activity and interest. For our purposes, we will use a simple,
standard deﬁnition of fairness: we will measure the throughput achieved by each node
over some time period, T , and say that an allocation with lower standard deviation is
“fairer” than one with higher standard deviation. Of course, we want the notion to work
properly when the number of nodes varies, so some normalization is needed. We will use
the following simpliﬁed fairness index:
i=1 xi )
( (10.1) where xi is the throughput achieved by node i and there are N backlogged nodes in all.
Clearly, 1/N ≤ F ≤ 1; F = 1/N implies that a single node gets all the throughput, while
F = 1 implies perfect fairness. We will consider fairness over both the long-term (many
thousands of “time slots”) and over the short term (tens of slots). It will turn out that in
the schemes we study, some schemes will achieve high utilization but poor fairness, and
that as we improve fairness, the overall utilization will drop.
Before diving into the protocols, let’s ﬁrst develop a simple abstraction for the shared
medium. This abstraction is a reasonable ﬁrst-order approximation of reality.
1. Time is divided into slots of equal length, τ .
2. Each node can send a packet only at the beginning of a slot.
3. All packets are of the same size, and equal to an integral multiple of the slot length. In
2 In this course, and in most, if not all, of the networking and communications world...
View Full Document
This document was uploaded on 02/26/2014 for the course CS 6.02 at MIT.
- Fall '13
- The Land