Concurrent multipath transfer (CMT) using the stream control transmission protocol (SCTP) can exploit multihomed devices to enhance data communications. While SCTP is a new transport layer protocol supporting multihomed end-points, CMT provides a framework so that transport layer resources are used efficiently and effectively when sending to the same destination with multiple IP addresses. In this paper, we present two techniques for modelling the expected throughput of a CMT session; while one is based on renewal theory, the other uses a Markov chain. As far as we know, ours is the first paper to model CMT whilst considering practical transport layer resources like a shared receive buffer (RBUF). A comparison of the models showed the Markov chain to be more accurate, but suffered from scalability issues. Alternatively, the renewal model was more cost effective, but also less accurate. We also applied our models to a new problem called congestion window management, where the size of each congestion window is reconfigured for optimal performance. Again, we compared two approaches: a dynamic method that makes decisions based on instantaneous throughput, and a static method that uses an integer linear program (ILP) to generate a global solution. Results showed the static method outperforming the dynamic approach by as much as 12 percent.

This paper proposes a new scheduling algorithm for concurrent multipath transfer (CMT) called the on-demand scheduler (ODS). Unlike previous algorithms, ODS waits for a transmission opportunity before making a scheduling decision. We showed ODS can outperform previous scheduling algorithms for CMT under various network scenarios. Moreover, this paper offers new insight into the effects of congestion and flow control on the receive buffer blocking problem associated with CMT.

SCTP is a new transport layer protocol extending the functionality of TCP with support for performance enhancing services such as multihoming and multistreaming. Unfortunately, an accurate mathematical model, able to approximate performance results, still eludes the research community. Previous models either fail to provide practicality, or are too simplified in their assumptions. This paper aims to circumvent these problems by developing an analytic model containing a higher degree of information. The result is a more accurate model that approximates the steady-state throughput of an SCTP session.

Telecommunication and grid computing applications demand high bandwidth data channels that offer guarantees with respect to service availability. Such applications include: remote surgery, remote experimentation, video on-demand, teleconferencing and bulk transfers. Furthermore, by forecasting traffic patterns internet service providers attempt to optimise network resources in order to lower operational costs during peak periods of bandwidth consumption. Advance reservation for wavelength division multiplexed networks can address some of these issues by reserving high volume communication channels (i.e. lightpaths) beforehand. The authors develop a mathematical model to solve the problem of scheduling lightpaths in advance. The optimal solution is presented as a mixed integer linear program with the assumption that all traffic is static and the network is centrally controlled. Furthermore, we have developed two novel meta-heuristics based on: 1) a greedy implementation (local search) and 2) simulated annealing. The meta-heuristics have shown to produce good approximate solutions in a reasonable amount of time.

Advance reservation is a topic that is rarely discussed within the domain of wavelength division multiplexed (WDM) networks. However, for many emerging applications in the telecommunication and/or grid computing industries, a demand for a high bandwidth communication channel as well as a guarantee on resource availability certainly exists. Such applications include: remote surgery, remote experimentation with teleobservation capabilities, teleconferencing, and bulk transfers. In this paper, we present a new model for reserving advance lightpath requests in a centralized system. This model attempts to “migrate,” i.e., move previously reserved lightpaths to candidate wavelengths in order to lower the system’s blocking probability. We have tailored different lightpath migration algorithms to address two specific network objectives: 1) minimize the number of hops a new request traverses after migration, and 2) minimize the number of migrated lightpaths. In terms of blocking probability, the lightpath migration algorithms show a significant improvement over the original advance lightpath reservation model.

This research aims to provide a network service for grid applications requiring high bandwidth with minimal to no delays while demanding strict scheduling constraints. More specifically, we are looking for efficient ways of scheduling advanced reservation communication requests at the network layer.