Simplicity, cost effectiveness, scalability, and economies of scale make Ethernet a popular choice for (a) local area networks (LAN), as well as for (b) storage area networks (SAN), and increasingly (c) metropolitan-area networks (MAN). Applications of Ethernet in the SAN and MAN arena elevate it from a LAN technology to a ubiquitous networking technology. With the expanded applicability of Ethernet there are certain adaptability issues which prompt rethinking of some of its architectural features. The Spanning-Tree based switching mechanism, considered to be very efficient at avoiding loops in LAN environments, is a performance bottleneck in the metro network context. Absence of an explicit switching path selection mechanism prohibits traffic engineering in metro networks. Prolonged spanning tree reconstruction periods after failures make Ethernet unsuitable to support critical applications. Lack of usage regulation mechanisms leads to insufficient isolation between different users, resulting in QoS problems.
Modern Ethernet switches support many advanced features which can be controlled through programmable interfaces. These features are VLAN tagging, rate limiting, and status monitoring. Conventionally, these features are mostly used to statically configure an Ethernet switch. This research proposes to use these features as dynamic control mechanisms in the context of metro and cluster networks to: (1) Maximize physical network link resources by enabling MPLS like traffic engineering, (2) Minimize failure recovery time, and (3) Enforce QoS requirements.
With these programmatic configurable control mechanisms, standard Ethernet switches can be used as effective building blocks for metropolitan-area Ethernet networks (MEN), storage-area networks (SAN), and computation cluster interconnects. We validated the usefulness of this new level of control over Ethernet switches with a MEN architecture that features multi-fold throughput gains and sub-second failure recovery time.
We discuss how a comprehensive resource management system can be devised using these mechanisms that can result in high performance and fault tolerant metro Ethernet, Storage, and Cluster networks. We also discuss how a network topology can be efficiently evolved by correlating the traffic profile characteristics of the end users and the traffic engineering required in the network. We describe a design methodology for Ethernet-based SAN fabrics utilizing this network evolution technique.
To this effect, we develop a network topology planning tool to minimize network infrastructure deployment cost. This topology planning work is specifically targeted towards providing automated tools to design Ethernet storage area network and cluster interconnect topologies with redundancy and fault-tolerance support.