CS 244 ’20: Reproducing Scafida: A Scale-free Network Inspired Datacenter Topology


In this paper, we reproduce the main findings of Scafida: A Scale-Free Network Inspired Data Center Architecture. The paper proposes and evaluates Scafida, a datacenter topology that breaks the symmetry traditionally found in state-of-the-art architectures, while allowing for arbitrary numbers of heterogenous equipment in the network and involving servers with more than one port in the routing process. The Scafida topology is inspired by scale-free networks (e.g. ordinary Barabasi-Albert topology), and attempts to exploit all the desirable properties that come along, namely short paths and high fault tolerance (path redundancy.) However, it imposes the additional constraint of limiting the degrees of nodes in the network, since neither switches nor servers can have arbitrary numbers of ports in practice. The original paper finds that such degree constraints have little negative effect on the resulting topology, and that Scafida has favorable properties, such as low mean path lengths, high throughput, and fault tolerance in the face of significant failures, all of
which are comparable to state-of-the-art topologies, despite its asymmetric structure.

Leave a comment