Correlation-Aware Virtual Machine Allocation for Energy-Efficient Datacenters

Abstract

Abstract-Server consolidation plays a key role to mitigate the continuous power increase of datacenters. The recent advent of scale-out applications (e.g., web search, MapReduce, etc.) necessitate the revisit of existing server consolidation solutions due to distinctively different characteristics compared to traditional high-performance computing (HPC), i.e., user interactive, latency critical, and operations on large data sets split across a number of servers. This paper presents a power saving solution for datacenters that especially targets the distinctive characteristics of the scale-out applications. More specifically, we take into account correlation information of core utilization among virtual machines (VMs) in server consolidation to lower actual peak server utilization. Then, we utilize this reduction to achieve further power savings by aggressively-yet-safely lowering the server operating voltage and frequency level. We have validated the effectiveness of the proposed solution using 1) multiple clusters of real-life scale-out application workloads based web search and 2) utilization traces obtained from real datacenter setups. According to our experiments, the proposed solution provides up to 13.7% power savings with up to 15.6% improvement of Quality-of-Service (QoS) compared to existing correlation-aware VM allocation schemes for datacenters. I. INTRODUCTIONThe soaring demand for computing has produced as collateral undesirable effect a surge in power consumption of servers and datacenters [1]. Server consolidation [3], which minimizes the number of active servers by packing workloads, or virtual machines (VMs) in a virtualized environment, into the minimal number of active servers, is one of the widely used techniques to reduce the power consumption of datacenters.However, in order to satisfy the performance demand of applications running on servers, server consolidation is usually conducted assuming the worst-case (or peak) utilization [2], [3]. Thus, in order to achieve power consumption without any significant Quality-of-Service (QoS) degradation, many works have presented aggressive consolidation schemes, which pack VMs based on off-peak (e.g., 90th/95th/99th percentile) of server utilization [4], [5]. Recently, correlation of resources utilization patterns among VMs are also exploited, such that, un-correlated VMs are co-located into a server to enable overprovision of VMs under negligible QoS degradation [6]- [9]. Nonetheless, these existing solutions are mostly designed for high-performance computing (HPC) applications and do not

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

“…Based on the observations and motivations above, we presented a server consolidation solution in [56]. First, to efficiently capture correlation information, they present a low-complexity measure for evaluating workload correlation among co-located VMs, and then, developed VM allocation algorithm.…”

“…In addition, Pearson's correlation is also partly inefficient because the value reflects correlation throughout the corresponding time interval because we only require the correlation at (off-)peak utilizations in VM placement. Equation (30) is presented in [56] as a new measure to quantify the correlation between two VMs to overcome the inefficiency of the conventional correlation metric.…”

“…1.2. To overcome the drawbacks of existing solutions, we [56] developed a power management solution for datacenters hosting scale-out application, especially targeting following distinctive workload characteristics of scale-out applications. We used a websearch application in CloudSuite [4] as a proxy to characterize the workload characteristics of scale-out applications.…”

“…Innovative DCIM support systems for datacenter management are thus needed. PMSM (i.e., Power Monitor System and Management) [44][45][46][47][48][49][50][51][52][53][54][55][56], developed at EPFL in cooperation with Credit Suisse [45], is an example of such an innovation.…”

Joint Computing and Electric Systems Optimization for Green Datacenters

“…Therefore, these approaches do not work well with non-stationary and fast-changing VM behaviors in particular for scale-out applications. In [22], a power-efficient solution is proposed based on the First-Fit-Decreasing heuristic to separate load correlated VMs especially targeting the characteristics of the scale-out applications. They also exploit server's Dynamic Voltage and Frequency Scaling (DVFS) techniques to achieve further energy savings.…”