Our increasing reliance on Internet-centric services and information technology has led to the proliferation of datacenters. The peak power consumption of these datacenters significantly impact their sustainability: both their recurring electricity bill (Op-ex) and one-time construction costs (Cap-ex). Existing pieces of work optimizing these cost are largely inadequate due to two main reasons, (i) they rely on overly conservative face-plate rating based estimates of datacenter peak needs, rendering them ineffective, and (ii) they employ demand throttling (DVFS) and/or demand shaping (load spreading/migration) techniques, both with heavy performance degrading implications. We address the first issue by developing fine-grained characterization of workload power/performance needs based on empirical power and resource-usage measurements. We further exploit the temporal/spatial variance in workload power requirements to develop statistical multiplexing based representation of aggregate datacenter power demand. Together, these measurement and statistical characterization allow us to make more informed power management decisions.

Towards addressing the second issue, we propose an entirely novel knob for re- ducing the peak power consumption which does not have any performance degrad- ing consequences, by exploiting the already existing energy buffer (eBuff) available in the form of UPS batteries in datacenters. Intuitively, eBuff stores energy in UPS batteries during “valleys”—periods of lower demand, which can be drained during “peaks”—periods of higher demand. UPS batteries are normally used as a fail-over mechanism to transition to captive power sources upon utility failure. Furthermore, frequent discharges can cause UPS batteries to fail prematurely. We conduct detailed analysis of battery operation to figure out feasible operating regions given such battery lifetime and datacenter availability concerns. Using insights learned from this analysis, we develop peak reduction techniques that combine the UPS battery knob with existing techniques for reducing datacenter power costs at minimal performance impact. Using an experimental platform, we find that eBuff can be used to realize up to 30% peak power reduction across a wide range of workload peaks/valleys, UPS provisioning, and application Service Level Agreement (SLA) constraints. Our cost-benefit analysis of investments in battery capacity and the resulting savings in Cap-ex and Op-ex suggests that datacenters are likely to gain substantially from additional procurement of battery capacity.