Reliable and deterministic network services
In a post Internet bubble economic environment the obvious lack of any disruptive new technologies that could give new impetus to network growth (as, for example, experienced after the appearance of the Web) lead to a trivialization of network services: Aside grass-roots, semi-subterranean applications like file sharing, entertainment content swapping and on-line gaming, it can be argued that network services are presently used in a manner not very different than a decade ago. In this respect, one of the factors that can be identified as limiting further development is the perceived lack of guarantees with regard to network services.
The articles comprised in this thesis address the subject of network service guarantees from two main perspectives.
In Part I we study network service guarantees from a reliability standpoint. In particular, we address the following issues:
Study what network infrastructure characteristics (e.g. data transmission method properties) are prone to inducing vulnerabilities in the overlaying network services.
Analyze the mechanisms and methods used at network protocol stack level for mitigating impairments like temporary traffic congestion or network element failures.
Propose mechanisms for fast network service level restoration in the wake of network elements failures.
In Part II of the thesis we address the topic of guarantees with respect to the network service performance level. In particular we study and propose mechanisms for guaranteeing deterministic upper bounds on worst-case queuing delays in general packet networks. Using network calculus concepts of traffic envelope and service curve we show how to deliver end-to-end queuing delay guarantees in connection-oriented networks carrying several traffic classes, with multiple packet sizes. Additionally, we measure the efficiency and performance level of such mechanisms and analyze inherent limitations of the proposed method.
queuing delay guarantees
network failure recovery