The Stream Control Transmission Protocol (SCTP) is a general-purpose transport protocol like TCP. However, unlike TCP, it provides support for so-called multi-homing. That is, endpoints (e.g. PCs, mobile devices, etc.) may be connected to multiple networks simultaneously. Clearly, if there are multiple paths available, it is quite useful to utilise all of them simultaneously, in order to maximise the payload throughput. An extension providing this feature for SCTP is denoted as Concurrent Multipath Transfer for SCTP (CMT-SCTP). Obviously, a CMT-SCTP connection should behave fairly to concurrent TCP flows. That is, if some of the paths share a common bottleneck link, the total bandwidth of a CMT-SCTP connection's paths should be equal to the bandwidth of a TCP flow on that link. Since the current Internet protocols do not provide a reliable mechanism for the detection of shared bottlenecks among paths, the CMT-SCTP congestion control just assumes that all paths are probably shared. Of course, in case of really disjoint paths, CMT-SCTP then acts less aggressively than necessary, leading to a lower performance than possible.
Recently, there have been some ideas of a statistical analysis of the data transfer behaviour, in order to detect shared bottlenecks for multi-path transmission. The algorithms have been implemented and tested in simulations, but an evaluation in the real Internet is still missing. To support the evaluation of multi-homed systems in real life, the Simula Research Laboratory is currently building up the NorNet testbed, a multi-site, multi-homed platform for network experiments in the Internet.
The subject of this masteroppgave will be to implement one of the approaches for shared bottleneck detection in a real system, in order to perform an evaluation within the NorNet testbed. This system will be based on the FreeBSD CMT-SCTP implementation and realised mostly in user-space, with possibly some small parts within the kernel itself. As reference, you will have access to the simulation implementation. Also, contact to the main FreeBSD CMT-SCTP developers is available. Particularly, the shared bottleneck detection implementation is also intended to be used in the context of the IETF standardisation process. Therefore, a further goal is to contribute this protocol extension to FreeBSD. That is, it is supposed that your work will in the future be included in many FreeBSD-based systems!
This project is *your* chance to get involved into CMT-SCTP development and worldwide IETF protocol standardisation as well as into international, top-level research at the Simula Research Laboratory in Fornebu! Are you interested in this challenge? Do not hesitate to contact us!
For more information please contact Michael Welzl (michawe@ifi.uio.no) and Thomas Dreibholz (dreibh@simula.no).
What you should know:
* You should have some programming experience, preferably in C.
What you will do:
* Implement shared bottleneck detection for CMT-SCTP under FreeBSD
* Run a performance evaluation within the NorNet testbed
* Contribute an interesting extension to FreeBSD CMT-SCTP.