PhD completed at the Centre for Communication Systems Research, University of Surrey.
Thesis for Doctor of Philosophy (PhD), June 2001.
Discusses satellite constellations with intersatellite links, some with double surface coverage or with MPLS.
Here, we examine networking and internetworking issues affecting satellite networking in complex satellite constellation networks, and determine what is needed in order to support services based on the TCP/IP suite well in satellite constellations.
We analyse constellation network topology. Its movement and effects on end-to-end delays experienced by network traffic travelling across the constellation are examined in detail. Analysis of the impact of cross-seam links upon delays experienced by traffic across star constellations shows that the use of cross-seam links is worthwhile.
We examine the effects of multi-path routing within the constellation upon TCP communication, and demonstrate the performance advantages of an intelligent flow-based approach to routing in the constellation network.
The desirability of implementing IP routing functionality in the space segment of the constellation is shown. The use of IP routing, to enable good support for IP QoS and IP multicast, is shown to be possible.
We present an approach to implementing IP multicast within the constellation, evaluating use of a core-based tree algorithm, and outline an architecture permitting IP routing of IP traffic in an ATM-based satellite constellation network, using MPLS.
Finally, we present and demonstrate the advantages of a novel method of managing path delay between ground terminals across a rosette constellation with intersatellite links, by using controlled handover to manage surface diversity to provide classes of service to network traffic.
|Thesis supervisor||Professor George Pavlou, University of Surrey|
(at University College London from January 2008).
|External examiner||Professor Jon Crowcroft, University College London|
(at University of Cambridge Computer Laboratory from October 2001).
|Internal examiner||Professor Rahim Tafazolli, University of Surrey|
(Centre for Communication Systems Research until August 2014, then Institute for Communication Systems).
Passed viva voce (defence) 21 February 2001;
completed minor corrections April 2001.
Letter of award dated 7 June 2001; graduation in Guildford Cathedral 30 November 2001.
Diagram tools: footprint generator, figure 1.1, p. 2.
SaVi, figures 1.2-1.8, pp. 3-10; figures 6.3-6.4, p. 151.
ns satellite plot scripts, figure 2.4 p. 24, figure 2.7 p. 27, figure 2.11 p. 35, figures 4.5-4.6 pp. 97-98, figures 4.19-4.21 pp. 112-114, figure 6.5 p. 152, figure 6.19 p. 162.
ns path delay script, ch. 2, ch. 6, Appendix 4.
Print errata: Contrary to the definition of RFC (p. xxi), not all RFCs are approved by the IESG.
TT&C (p. xxii) is often known as Telemetry, Tracking and Command.
Fraise00 (p. 4) should refer to Fraiseetal00 (p. 180). Seo88 (p. 188) was, of course, published in 1988.
RFC896 (p. 184) is referenced - but why? Interaction between Nagle and delacks is not touched on...
Section 2.1.3 (p. 28) describes the Walker notation as N/P/p. Walker did adopt a revised notation of NP/P/p, which can be confused with Ballard's notation. '...especially when m is a fractional' should read 'especially when the phase offset is an unusual harmonic fraction'.
Chapeter 3 is not aware that real IP networking devices do load balancing by hashing on source and destination addresses, so that TCP flows are not reordered by round-robin multipath forwarding.
Section 4.2.2 (p. 82) mentions a millisecond counter. IPv4's time-to-live would rely on a second counter of unspecified accuracy. See p. 30 of RFC791 for the original specification, which includes decrementing by a minimum of one after every hop.
Section 4.8.4 (p. 110) begins by correctly referring to the longitude of the seam, and should go on to say that each core position is labelled with the longitude at which it becomes the position. Figures 4.19 to 4.21 should also refer to longitude rather than latitude, as is clear from their map axes. (longitude becomes longitude' once coordinates are transformed, and is considered as "latitude" or as angle from the centre of the seamed cylinder once in cylindrical coordinates.)
Section 6.3.1 (p. 152) describes an 'occasional (but small) gap in coverage' in the Celestri simulation. This was due to adding the phase offset when we should have been subtracting it -- but you still need broader beams to get double surface coverage. The SaVi Celestri simulation has been adjusted.