Provisioning and Harnessing Energy Storage in Datacenters

Open Access
Wang, Di
Graduate Program:
Computer Science and Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
April 30, 2014
Committee Members:
  • Anand Sivasubramaniam, Dissertation Advisor
  • Anand Sivasubramaniam, Committee Chair
  • Bhuvan Urgaonkar, Committee Member
  • Chitaranjan Das, Committee Member
  • Hosam Kadry Fathy, Committee Member
  • Aman Kansal, Special Member
  • Lee David Coraor, Committee Member
  • Datacenters
  • Power Management
  • Energy Storage
  • Batteries
With terawatts expended in powering today's datacenters, their power consumption poses serious economic, societal and environmental concerns. A large datacenter spends millions of dollars in yearly operational expenditures (op-ex) paying its electricity bills. An even larger capital expenditure (cap-ex) goes into provisioning its power infrastructure, to accommodate the peak power draw, even if this draw is never or rarely sustained. With consumers demanding more for less, extracting the maximum value out of every provisioned and consumed watt in these datacenters is critical to profitability and sustenance. There have been proposals to underprovision the power infrastructure, sizing it to handle a high percentile of the power draw rather than occasional high peaks. Demand Response (DR) is then employed to address these high peaks at runtime by re-shaping, deferring or modulating demand temporally and spatially. Until now, datacenter DR has primarily used computing knobs such as consolidation, scheduling, migration, and power state modulation. Recently, Energy Storage Devices (ESDs) have been proposed to provide a complementary alternative to these knobs. However, there are multiple issues and challenges in leveraging these ESDs for datacenter DR that require an in-depth study in the datacenter context, which is one major intent of this dissertation. On the other hand, the primary usage of energy storage in today's datacenters, in the form of Uninterrupted Power Supply (UPS) devices, is to handle power outages. A significant portion of the datacenter power infrastructure's cap-ex goes into the backup power equipment - Diesel Generators (DGs) and UPSes - needed to sustain operation during utility power outages. However, with only a few power outages per year and the majority outages being shorter than a few minutes, it implies opportunities for underprovisioining backup infrastructure for cost savings. However, embarking on such underprovisioning mandates studying several ramifications - the resulting cost savings, the lower availability, and the performance and state loss consequences on individual applications - concurrently, which is another major focus of this dissertation. The goal of this dissertation is to develop solutions for provisioning and harnessing ESDs to (i) improve datacenter demand response capabilities, and (ii) optimize datacenter backup power infrastructure cost via modeling, simulation and experimentation with datacenter workloads. The dissertation consists of two major parts. In the first part, we mainly focus on how to leverage ESDs for enhancing datacenter demand response capabilities. We first develop three different methodologies for provisioning and control energy storage: (1) heuristics based on real datacenter power demand peak and valley characteristics for easy and quick ESD provisioning; (2) an analytical model which captures the important peak/valley attributes and ESD properties to compare different ESD cost-effectiveness in shaving different kinds of power demands; (3) a generalized optimization platform for ESD provisioning, placement and control. Then, we design and implement a system software for managing these ESDs. With a 2-level power hierarchy prototype, we evaluate the impacts and benefits of different mechanisms and policies in harnessing ESDs using datacenter applications. In the second part, we explore the datacenter backup power infrastructure design space, considering cost, availability, performance and application consequences of underprovisioning the backup infrastructure. We present a framework to quantify the cost of backup capacity that is provisioned, and implement techniques leveraging existing software and hardware mechanisms to provide as seamless an operation as possible for an application during a power outage. In the evaluation, we show that Diesel Generators can be completely removed in many cases, and compensated with additional battery energy capacity in the UPS units to achieve desirable cost-performance-availability tradeoff.