Ethernet label spaces dependency on network topology

Abstract

SUMMARYLabel space consumption has been studied in label-based forwarding architectures such as multi-protocol label switching (MPLS) to reduce forwarding table sizes and lookup complexity, to simplify network management, and to limit operational expenditures. On the other hand, nowadays label-based forwarding is also considered in the context of carrier class Ethernet architectures. Given that these architectures may use different label scopes (domain-wide vs. local) and size spaces (depending on their encoding), there is a need to analyse and compare the properties of their respective label spaces, particularly in terms of scalability. In this paper, the impact of topology characteristics on label space consumption is studied. The dependency on factors such as the number of nodes, the topology node degree and the network structure is determined (both analytically and via experimentation), allowing a detailed comparison between the properties of the evaluated label spaces. Proposed techniques that can be applied to improve label space usage for label-based Ethernet forwarding architectures are considered. The proposed methodology and results can serve as guidelines for the design of future label based forwarding architectures.

Torii: multipath distributed Ethernet fabric protocol for data centres with zero‐loss path repair

“…Once we have explained the applicability of Torii to different topologies, in this section, we address the scalability of Torii by comparing it directly with PortLand, which is already a quite scalable data centre proposal. Scalability is an important issue in Ethernet networks, as described in . For the comparison, we will use different hierarchical topologies, which are applicable to Torii and at the same time define common examples of data centre topologies of two, three and four levels of hierarchy.…”