Demand for datacenter capacity is constantly increasing worldwide, primarily driven by cloud computing. As a result, large cloud providers are currently deploying costly, multi-gigawatt facilities annually. To help reduce costs, this thesis is dedicated to cost-eﬃcient methods for provisioning datacenter infrastructure resources.
First, we propose a method to reduce costs by under-provisioning the cooling infrastructure of datacenters via deploying cheaper cooling technologies or capacities. Our results suggest that our proposed method can reduce the cooling capital costs by up to 55%. We then turn our attention to the impact of cheaper cooling technologies to server component reliability. We draw interesting observations, including 1) disk failures increase signiﬁcantly when datacenters operate at high relative humidity and 2) though higher relative humidity increases failure rates, the infrastructure savings far outweigh the cost of the extra component failures. Finally, we propose a framework that generates cost-eﬃcient placements and provisionings of datacenter ﬂeets which are deployed at the edge of the internet. For interactive applications with tight performance constraints, our framework suggests edge fleet provisionings that cover 1.84x more population than a real Microsoft hub ﬂeet with just 20% higher per customer costs.