There is a popular mis-understanding concerning the performance of TCP (used by most Internet applications) over satellite. This is fuelled by various misconceptions which have become widespread in recent years. The mis-understanding stem from three areas:
(i) Many people have experience with implementations which are now out of date. Research papers published in the later 1980's and early 1990's identified problems with TCP, most of which have now been resolved.
(ii) Many reported experiments have used mis-configured TCP protocol stacks. Until recently few protocol stacks have been configured suitably for satellite operation.
(iii) Unfortunately, many researchers still fail to understand the way in which TCP actually operates. This has lead to a lot of bad advice. The IETF has recently published two documents to try and address these issues.
The result of all this, is that many people are left with fear, uncertainty and doubt concerning whether TCP will actually operate over a satellite link. In fact, this doubt is mostly unfounded, as many satellite service users can testify. TCP was designed to work over satellites. The original goal of the research project that created TCP was to link a pilot satellite network (SATNET) with the terrestrial Internet (ARPANET). Although there is no practical limit to TCP's throughput (the maximum theoretical throughput is 1.5 Gbps - faster than any satellite link), there a number of significant issues which do directly impact the performance which real users are likely to achieve. To understand things more clearly, it is first necessary to examine how TCP has evolved.
SPECIFIC ISSUES THAT IMPACT THE SATELLITE SERVICE
The main distinguishable characteristics of a satellite link compared to a terrestrial network which effect the performance of TCP are:
(i) Effect of Satellite Link Errors
(ii) Satellite Propagation Delay
(iii) Bandwidth and Path Asymmetry
(iv) Channel Access and Network Interactions
The ERG Communications group continues to study these issues and to develop techniques to reduce their impact.
The IETF has recently published two working documents to try and explain these issues and counter the misconceptions described previously.