Free cooling-aware dynamic power management for green datacenters

Abstract

Abstract-Free cooling, i.e., directly using outside cold air and/or water to cool down datacenters, can provide significant power savings of datacenters. However, due to the limited cooling capability, which is tightly coupled with climate conditions, free cooling is currently used only in limited locations (e.g., North Europe) and periods of the year. Moreover, the applicability of free cooling is further restricted along with the conservative assumption on workload characteristics and the virtual machine (VM) consolidation technique as they require to provision higher cooling capability. This paper presents a dynamic power management scheme, which extends the applicability of free cooling by judiciously consolidating VMs exploiting time-varying workload characteristics of datacenter as well as climate conditions, in order to minimize the power consumption of the entire datacenter while satisfying service-level agreement (SLA) requirements. Additionally, we propose the use of a receding horizon control scheme in order to prevent frequent cooling mode transitions. Experimental results show that the proposed solution provides up to 25.7% power savings compared to conventional free cooling decision schemes, which uses free cooling only when the outside temperature is lower than predefined threshold temperature.

Power-Thermal Modeling and Control of Energy-Efficient Servers and Datacenters

“…2.2, we need to revisit existing VM placement solutions [60][61][62][63][64][65] as it further reduces the chance of using free cooling as the solutions requires higher cooling capability due to the higher operating temperature of active servers. Motivated by this observation, Kim et al present a joint power and thermal optimization solution for datacenters equipped with hybrid cooling architecutre to achieve further power savings while satisfying service-level agreement (SLA) requirements by extending the usability of free cooling for datacenters having a hybrid cooling architecture [43]. Figure 28 illustrates the solution overview explained in this section.…”

“…The problem is a well-known bin-packing problem [43]. In order to reduce the solution complexity within negligible solution quality degradation, Kim et al present a heuristic based on a First-Fit-Decreasing where it first manipulates VMs having the highest utilization among unallocated VMs.…”

“…Fig. 8 Datacenter cooling architecture [43] Despite the promising advantages on cooling-energy efficiency, the fundamental issue of free cooling is its limited applicability, as it can only be used in a very limited set of geographical locations because the cooling capability is tightly coupled with climate condition (e.g., temperature and humidity). Thus, it suffers from wide variations of cooling efficiency during the year, which translates in significantly high computing systems failure rates [25].…”

“…Power usage effectiveness (PUE) in electrical and free cooling as power consumption of server varies [43] Finally, the temperature of the server room, T room , depends on CRAH efficiency, CRAH , which is defined as follows [43]:…”

Stochastic MPC Framework for Controlling the Average Constraint Violation

“…Other examples include energy-efficient datacenter cooling subject to the constraints on the number of delayed queries per unit of time (see, e.g., [17]), or performance (e.g., power output) maximization of a machine subject to fatigue constraints (see [9] for a concrete example of wind-turbine control).…”