On the sustainability of PEPs for satellite Internet access

TCP protocols such as New Reno suffer from very poor performances in satcom. In reaction, several solutions have been deployed to improve the global performance. PEP (Proxy Enhancement Performance) is the most widely adopted. However, this improvement has a cost, and using PEPs adds new constraints to face. Recently, many works not specific to the satellite scope have been done to develop new TCP protocols leading to new TCP releases such as Compound and Cubic TCP. This article investigates the behavior of those new protocols in satellite communication. The main objective of our work is to determine if using specific satcom techniques is still a sustainable approach. I. Introduction Satellite communications are an attractive way to communicate efficiently with large or unreachable areas. As a consequence, they deserve to be part of the global communication network. However, satcom systems have to deal with some very challenging and restrictive constraints impacting on different level of the network protocol stack. Among this, for the particular scope of Transport Protocol, long Round Trip Times (about 600 ms for GEO system) are strongly damaging the global behavior. Indeed, they impair directly some congestion avoidance mechanisms such as slow start that are an essential part of reliable transport protocol (e.g. TCP), and so decrease the throughput. Moreover, satellite telecommunication system is a small niche of the global communication system. A new transport protocol version dedicated to satellite is then unthinkable, especially if it could weaken the overall performance of the legacy protocol. Considering those impairments, Performance Enhancement Proxy (PEP) solution has been developed and is currently the most widely adopted architecture for satellite links. PEPs are a pragmatic solution using a transport level proxy to separate the satellite segment from the others (notably using splitting and spoofing methods). The segment becomes transparent for the rest of the network, and so the global behavior of the transport protocol is almost unchanged for the extremities, which means better throughput and performance. 4 Moreover, the end–to–end implementation remains unaltered. During the last few years, new elements became apparent which are likely to cast doubt on the PEP solutions. Thus, some limits, mostly imputable to the break of the end–to–end semantics have been discovered, and notably, that PEPs behave badly under network mobility scenarios or some security context (e.g. IPsec tunnel). Alongside this, new TCPs like Cubic and Compound have been developed for a non specific satellite use. And they seem to behave well in an hybrid scenario and already better than end–to–end specific satellite transport protocol such as TCP Hybla. 14 This study aims to determine the usefulness of transport protocol PEP in two different scopes. To do that, emulation of TCP New Reno and modern TCP in an hybrid scenario with and without PEP are used. First of all, for a short-term optimization purpose, the aim is to define if modern TCPs perform well enough without PEP (throughput, fairness, ...) to question the usefulness of current PEPs. Secondly, for a sustainable purpose, this study tries to determine if PEPs can still remain performant