The Case for Medium-Sized Regional Data Centres

Abstract

Cloud computing is widely associated with major
capital investment in mega data centres, housing expensive blade
servers and storage area networks. In this paper we argue that a
modular approach to building local or regional data centres using
commodity hardware and open source hardware can produce a
cost effective solution that better addresses the goals of cloud
computing, and provides a scalable architecture that meets the
service requirements of a high quality data centre.
In support of this goal, we provide data that supports three
research hypotheses:
1. that central processor unit (CPU) resources are not
normally limiting;
2. that disk I/O transactions (TPS) are more often
limiting, but this can be mitigated by maximizing the
TPS-CPU ratio;
3. that customer CPU loads are generally static and
small.
Our results indicate that the modular, commodity hardware
based architecture is near optimal. This is a very significant
result, as it opens the door to alternative business models for the
provision of data centres that significantly reduce the need for
major up-front capital investment.