work. That is, we assume traffic values typical of
the ILUs. The goal of satellite Internet access
design is to provide a service to ALUs which will
transform each ALU into an ILU.
mechanism for access must utilize the satellite
transponder resource in an efficient manner,
since it is expensive.
The growth of
the Internet as an
efficient and
effective tool for
information
These considerations lead to the conclusion
that high-quality access to the Internet for the
ALUs of interest in this article requires a con-
nection-free wideband access architecture. A
connection-free architecture for the multiple
access channel completely eliminates any con-
nection-related overhead in the multiple access
channel. At the same time, a connection-free
wideband architecture eliminates any latency
caused by the connection process.
Eliminating the inefficiencies generated by a
connection-oriented architecture in the presence
of bursty traffic is an obvious advantage of con-
nection-free operation. Another less obvious
advantage of wideband connection-free opera-
tion of the multiple access channel is the scala-
bility and the fungible nature of data flow in such
a channel.
When a connection-oriented architecture is
used, the question arises as to how large to make
the connections or channels assigned in the net-
work. If the connections established are at a low
data rate, the transmission of data in the chan-
nels assigned may be too slow, while if the con-
nections are established at a high data rate, the
utilization of the satellite resource may not be
efficient. Some reasonable compromise must be
made, but this “one size fits all” approach can
lead to further inefficiencies when the channel
traffic does not fit the choice made.
Since the amount of traffic in the broadcast
channel is typically much greater than the amount
of traffic in the multiple access channel, we find it
convenient to first characterize the average
amount of traffic per user per day in the broadcast
channel. Without trying to be too precise on this
number we take the value of 2 Mbytes/day/user as
typical of the traffic to the user in the networks of
interest. Certainly there will be users who receive
much more than 2 Mbytes/day, and there are days
when almost any user of the Internet will receive
more than this value. Nevertheless, we take this
number as a reasonable guide to the average
amount of traffic delivered to a typical user over a
long period of time.
In order to specify the average traffic in the
multiple access channel, we assume a value for
the asymmetry of the access network; that is, a
value for the ratio of the average amount of data
sent in the broadcast channel to the average
amount of data sent in the multiple access chan-
nel. Again, this number is subject to a great deal
of uncertainty and variation among different
users as well as over time with a single user. But
with the same level of confidence and precision
with which we assumed the average broadcast
channel traffic in the preceding paragraph, we
assume the asymmetry of typical Web and e-mail
traffic to be about six to one.
retrieval is based
upon the shared
utilization of
high-capacity
information
resources. On the
Internet, servers
are shared,
switches are
shared and the
Internet
backbone is
shared among
large numbers
of users.
The characterization of the average traffic in
the multiple access channel and the asymmetry
of the traffic in the two channels provides a nec-
essary but incomplete description of the network
traffic. The traffic in both channels, of course, is
bursty, but the bursty nature of the traffic in the
broadcast channel is not a critical factor in
determining the performance of that channel.
Since the broadcast channel provides a one-to-
many communication link, the bursty nature of
the traffic in the broadcast channel is smoothed
by buffering in the hub transmitter. In the many-
to-one multiple access channel, however, the
bursty nature of the traffic flow is smoothed to a
much more limited degree.
In a connection-free architecture the data
flowing within the channel is fungible. Thus, for
a wideband connection-free channel with a
throughput of 1 Mb/s, the traffic can be provid-
ed by 10 users each with an average data rate
of 100,000 b/s , by 100 users each with an aver-
age data rate of 10,000 b/s , by 100,000 users
each with an average data rate of 10 b/s, or by
any combination of users as long as the aggre-
gate average data rate does not exceed the
maximum throughput of 1 Mb/s. The value of
this flexibility for the case of Internet traffic is
important precisely because of the unknown
nature of the traffic in this rapidly changing
medium. It is difficult to specify the nature of
the Internet traffic of today. The specification
of the Internet traffic of tomorrow is much
more speculative, so a scalable, fungible archi-
tecture insensitive to arbitrary traffic assump-
tions is highly desirable.
CONNECTED ACCESS AND
CONNECTION-FREE ACCESS
The growth of the Internet as an efficient and
effective tool for information retrieval is based
on the shared utilization of high-capacity infor-
mation resources. On the Internet, servers are
shared, switches are shared, and the Internet
backbone is shared among large numbers of
users. In the case of ILUs, access to the Internet
is also shared, and the sharing process makes
possible high-speed access at a reasonable shared
cost. By extending the sharing of resources from
the Internet to the ALUs, it is possible to pro-
vide the same level of high-performance yet
affordable service to these users.
PROTOCOLS
When the communication channel resource is a
limiting cost factor in system design, it is neces-
sary to examine the suitability of how that
resource is used. In the case of satellite channels
used for Internet access, a major component of
that usage is the transmission of protocol infor-
mation within the satellite channels. Of course,
some protocol information is absolutely neces-
sary for the proper functioning of the communi-
cation path. But in the case of Internet traffic,
based in large part on the TCP/IP protocols
designed in the 1970s, the time has come to
reexamine the suitability of those protocols. This
reexamination should consider the possibility of
enhancing TCP to take into account the proper-
For ALUs, which are the focus of this article,
it is necessary to find a sharing mechanism for
bursty Internet traffic which will permit the high-
speed access required for the Internet applica-
tions of today. At the same time, such a sharing
62
IEEE Communications Magazine • July 2000