dth=100%>
Yahoo! is not affiliated with the authors of this page or responsible for its content.
Transport Protocols for Internet-Compatible Satellite Networks
TO APPEAR, IEEE JOURNAL ON SELECTED AREAS OF COMMUNICATIONS, 1999
1
Transport Protocols for Internet-Compatible
Satellite Networks
Thomas R. Henderson, Student Member, IEEE, and Randy H. Katz, Fellow, IEEE
Abstract We address the question of how well end-to-end transport
connections perform in a satellite environment composed of one or more
satellites in geostationary orbit (GEO) or low-altitude earth orbit (LEO), in
which the connection may traverse a portion of the wired Internet. We rst
summarize the various ways in which latency and asymmetry can impair
the performance of the Internets Transmission Control Protocol (TCP),
and discuss extensions to standard TCP that alleviate some of these perfor-
mance problems. Through analysis, simulation, and experiments, we quan-
tify the performance of state-of-the-art TCP implementations in a satellite
environment. A key part of the experimental method is the use of trafc
models empirically derived from Internet trafc traces. We identify those
TCP implementations that can be expected to perform reasonably well, and
those that can suffer serious performance degradation. An important re-
sult is that, even with the best satellite-optimized TCP implementations,
moderate levels of congestion in the wide-area Internet can seriously de-
grade performance for satellite connections. For scenarios in which TCP
performance is poor, we investigate the potential improvement of using a
satellite gateway, proxy, or Web cache to split transport connections in
a manner transparent to end users. Finally, we describe a new transport
protocol for use internally within a satellite network or as part of a split
connection. This protocol, which we call the Satellite Transport Protocol
(STP), is optimized for challenging network impairments such as high la-
tency, asymmetry, and high error rates. Among its chief benets are up to
an order of magnitude reduction in the bandwidth used in the reverse path,
as compared to standard TCP, when conducting large le transfers. This is
a particularly important attribute for the kind of asymmetric connectivity
likely to dominate satellite-based Internet access.
KeywordsInternet, Transport protocols, satellite communication, TCP.
I. I
NTRODUCTION
 
EVERAL companies (e.g., Alcatel, Hughes, Teledesic) have
recently announced plans to build large satellite systems to
provide commercial broadband data services distinct from nar-
rowband voice services. These systems are expected to offer
Internet access to remote locations and to support virtual private
networks for widely scattered locations. However, the perfor-
mance of data communications protocols and applications over
such future systems is the subject of heated debate in the re-
search community. Nowhere has this debate raged more than in
discussions regarding the transport-level protocol in the Internet
TCP/IP protocol suite (namely, the Transmission Control Proto-
col [1]). Some researchers insist that TCP will work suitably in
a satellite environment, while others have suggested satellite-
specic protocol options for improved performance, and still
others claim that TCP cannot work effectively over satellite
channels. There is, however, no disagreement in that the large
latencies, bandwidth and path asymmetries, and occasionally
high error rates on satellite channels provide TCP with a chal-
lenging environment in which to operate.
In this paper, we evaluate just how well TCP performs in
Manuscript received February 15, 1998; revised September 10, 1998. The
authors are with the Electrical Engineering and Computer Science Department
at the University of California, Berkeley. T. Henderson was also supported by
HRL Laboratories, Malibu, CA, for a portion of this work.
E-mail:
¡
tomh, randy
¢
@cs.berkeley.edu.
a satellite environment composed of one or more satellites in
geostationary orbit (GEO) or low-altitude earth orbit (LEO), in
which the end-to-end connection may traverse a portion of the
wired Internet. We rst discuss our assumptions concerning
future broadband satellite systems that plan to provide direct-
to-user Internet access, focusing on characteristics that impact
transport layer protocol performance. In Section 3, we describe
the various ways in which latency and asymmetry can impair the
performance of TCP, and discuss extensions to standard TCP
that alleviate some of these performance problems. Through
analysis, simulation, and experiments described in Sections 4
and 5, we quantify the performance of state-of-the-art TCP in a
satellite environment, both for large le transfers and short Web
transactions. A key part of our experimental method is the use
of trafc models empirically derived from Internet trafc traces.
We identify scenarios where TCP can be expected to perform
reasonably well, and where it can suffer serious performance
degradation due to either suboptimal protocol conguration or
congestion in the wide-area Internet. For the cases in which
performance is poor, we next investigate in Section 6 the im-
provements that can be gained by using a transport gateway to
split the end-to-end connection in a manner transparent to the
end user. Finally, in Section 7 we describe a new transport pro-
tocol for use within a satellite subnetwork or on the satellite side
of a split connection. This protocol, which we call the Satellite
Transport Protocol (STP), is optimized for challenging network
impairments experienced by satellite networks such as high la-
tency, bandwidth and path asymmetry, and high error rates.
The following are our main contributions:
£
Previous studies of TCP performance over satellite channels
have focused on the large le transfer performance of a sin-
gle connection in isolation, often on channels with high bit er-
ror rates. Our data indicates that, despite the use of satellite-
optimized TCP implementations on clean satellite channels, the
presence of other competing TCP connections in the wide-area
Internet can dominate the satellite connections performance.
We also illustrate how subtle implementation details can have
a major effect on TCP performance over satellite channels.
£
We quantify the effects of TCP latency on small data trans-
fers by performing analysis and experiments based on traces of
HTTP connections (the Hypertext Transfer Protocol [2], used
for Web browsing), and evaluate the relative merits of proposed
TCP options that reduce the latency of short HTTP connections.
£
We describe the design of STP, an adaptation of a reliable
ATM link layer protocol known as SSCOP, to provide trans-
port service in a connectionless network environment. Besides
being efcient and resilient to loss in the forward direction of
data transfer, the chief advantage of this protocol relative to
satellite-optimized TCP is a substantial reduction in the band-
width needed in the reverse channel. TO APPEAR, IEEE JOURNAL ON SELECTED AREAS OF COMMUNICATIONS, 1999
2
Wide-area
host
Internet
subnet
servers
GEO or (multihop) LEO satellite network
Satellite
clients
Satellite
Other
Other
Satellite
hosts server
ooooooooo
ooooooooo
ooooooooo
ooooooooo
ooooooooo
ooooooooo
ooooooooo
ooooooooo
ooooooooo
ooooooooo
ooooooooo
ooooooooo
ooooooooo
ooooooooo
¤
¤
¤
¤
¤
¤
¤
¤
¤
¤
¤
¤
¥
¥
¥
¥
¥
¥
¥
¥
¥
¦
¦
¦
¦
¦
¦
¦
¦
¦
¦
¦
¦
§
§
§
§
§
§
§
§
§
Fig. 1. Example of a future satellite network in which a satellite-based host communicates with a server in the Internet
II. T
RANSPORT ENVIRONMENT OF FUTURE SATELLITE
SYSTEMS
Our assumptions about future satellite network characteristics
are shaped by projections of future commercial systems (e.g.,
Teledesic [3], Spaceway [4]). These future systems will of-
fer Internet connections at up to broadband (tens of Mb/s) data
rates via networks of LEO or GEO satellites (or hybrid constel-
lations). Users may contact other hosts in either the satellite net-
work or the wide-area Internet. In general, we have considered
an architecture based on packet switching that is fully compati-
ble with the TCP/IP protocol suite. We also were primarily con-
cerned with architectures that scale to serving many thousands
of users (e.g., direct-to-user services rather than carrier trunk-
ing). Figure 1 illustrates the general topology, in which users
or small networks access the wide-area Internet via the satellite
system.
The main characteristics of the end-to-end path that affect
transport protocol performance are latency, bandwidth, packet
loss due to congestion, and losses due to transmission errors.
If part of the path includes a satellite channel, these parameters
can vary substantially from those found on wired networks. We
make the following assumptions about the performance charac-
teristics of future systems:
£
Latency:
The three main components of latency are prop-
agation delay, transmission delay, and queueing delay. In the
broadband satellite case, the dominant portion is expected to be
the propagation delay. For connections traversing GEO links,
the one-way propagation delay is typically on the order of 270
ms, and may be more depending on the presence of interleavers
for forward error correction. Variations in propagation delay
for G