Georgios Varsamopoulos

Energy-Efficient Thermal-Aware Task Scheduling for Homogeneous High-Performance Computing Data Centers: A Cyber-Physical Approach

High-performance computing data centers have been rapidly growing, both in number and size. Thermal management of data centers can address dominant problems associated with cooling such as the recirculation of hot air from the equipment outlets to their inlets and the appearance of hot spots. In this paper, we show through formalization that minimizing the peak inlet temperature allows for the lowest cooling power needs. Using a low-complexity linear heat recirculation model, we define the problem of minimizing the peak inlet temperature within a data center through task assignment (MPIT-TA), consequently leading to minimal cooling-requirement. We also provide two methods to solve the formulation: Xlnt-GA, which uses a genetic algorithm, and Xlnt-SQP, which uses sequential quadratic programming. Results from small-scale data center simulations show that solving the formulation leads to an inlet temperature distribution that, compared to other approaches, is 2 degC to 5 degC lower and achieves about 20 to 30 percent cooling energy savings at common data center utilization rates. Moreover, our algorithms consistently outperform the minimize heat recirculation algorithm, a recirculation-reducing task placement algorithm in the literature.

Thermal-aware task scheduling for data centers through minimizing heat recirculation

The thermal environment of data centers plays a significant role in affecting the energy efficiency and the reliability of data center operation. A dominant problem associated with cooling data centers is the recirculation of hot air from the equipment outlets to their inlets, causing the appearance of hot spots and an uneven inlet temperature distribution. Heat is generated due to the execution of tasks, and it varies according to the power profile of a task. We are looking into the prospect of assigning the incoming tasks around the data center in such a way so as to make the inlet temperatures as even as possible; this will allow for considerable cooling power savings. Based on our previous research work on characterizing the heat recirculation in terms of cross-interference coefficients, we propose a task scheduling algorithm for homogeneous data centers, called XInt, that minimizes the inlet temperatures, and leads to minimal heat recirculation and minimal cooling energy cost for data center operation. We verify, through both theoretical formalization and simulation, that minimizing heat recirculation will result in the best cooling energy efficiency. XInt leads to an inlet temperature distribution that is 2degC to 5degC lower than other approaches, and achieves about 20%-30% energy savings at moderate data center utilization rates. XInt also consistently achieves the best energy efficiency compared to another recirculation minimized algorithm, MinHR.

Cooling-aware and thermal-aware workload placement for green HPC data centers

High Performance Computing (HPC) data centers are becoming increasingly dense; the associated power-density and energy consumption of their operation is increasing. Up to half of the total energy is attributed to cooling the data center; greening the data center operations to reduce both computing and cooling energy is imperative. To this effect: i) the Energy Inefficiency Ratio of SPatial job scheduling (a.k.a. job placement) algorithms, also referred as SP-EIR, is analyzed by comparing the total (computing + cooling) energy consumption incurred by the algorithms with the minimum possible energy consumption, while assuming that the job start times are already decided to meet the Service Level Agreements (SLAs); and ii) a coordinated cooling-aware job placement and cooling management algorithm, Highest Thermostat Setting (HTS), is developed. HTS is aware of dynamic behavior of the Computer Room Air Conditioner (CRAC) units and places the jobs in a way to reduce the cooling demands from the CRACs. Dynamic updates of the CRAC thermostat settings based on the cooling demands can enable a reduction in energy consumption. Simulation results based on power measurements and job traces from the ASU HPC data center show that: i) HTS reduces the SP-EIR by 15% compared to LRH, a thermal-aware spatial scheduling algorithm; and ii) in conjunction with FCFS-Backfill, HTS increases the throughput per unit energy by 6.89% and 5.56%, respectively, over LRH and MTDP (an energy-effcient spatial scheduling algorithm with server consolidation).

Thermal aware server provisioning and workload distribution for internet data centers

With the increasing popularity of Internet-based information retrieval and cloud computing, saving energy in Internet data centers (a.k.a. hosting centers, server farms) is of increasing importance. Current research approaches are based on dynamically adjusting the active server set in order to turn off a portion of the servers and save energy without compromising the quality of service; the workload is then distributed, conventionally equally (i.e. balanced), across the active servers. Although there is ample work that demonstrates energy savings through dynamic server provisioning, there is little work on thermal-aware server provisioning. This paper provides a formulation of the thermal aware active server set provisioning (TASP), in a nonlinear minimax binary integer programming form, and a series of heuristic approaches to solving them, namely MiniMax, bb-sLRH, CP-sLRH and sLRH. Furthermore, it introduces thermal-aware workload distribution (TAWD) among the active servers. The proposed heuristics are evaluated using a thermal model of the ASU HPCI data center, while the request traffic is based on real web traces of the 1998 FIFA World Cup as well as the SPECweb2009 suite. The TASP heuristics are found to outperform a power-aware-only server set selection scheme (CPSP), by up to 9.3% for the simulated scenario. The order of achieved energy efficiency is: MiniMax (9.3% savings), CP-sLRH (9.2%), bb-sLRH (8.6%), sLRH (5.8%), compared to CPSP.

Trends and effects of energy proportionality on server provisioning in data centers

Cloud is the state-of-the-art back-end infrastructure for most large-scale web services. This paper studies what effect energy proportionality has on the energy savings of cloud data center management, under various equipment compositions and power densities. Our findings show that although it is a common expectation that improved energy proportionality should diminish the benefits of power managements server provisioning, this is not true in all cases. Results show that equipping server provisioning with thermal awareness can keep it as a useful technique when the data center exhibits power consumption heterogeneity and non-uniform heat recirculation phenomena.

TACOMA: Server and workload management in internet data centers considering cooling-computing power trade-off and energy proportionality

A two-tier Internet data center management scheme, TACOMA, with thermal-aware server provisioning (TASP) in one tier, and thermal-aware workload distribution (TAWD) in the other is proposed. TASP and TAWD coordinate to maximize the energy savings by leveraging the workload dynamics, at coarse and fine time scale, respectively. TACOMA is aware of the QoS constraints, the energy proportionality of servers, and the potential trade-off between cooling and computing power. The obtained energy savings are a combination of suspending idle servers, using servers at their peak efficiency, and avoiding heat recirculation.

GDCSim: A tool for analyzing Green Data Center design and resource management techniques

Energy consumption in data centers can be reduced by efficient design of the data centers and efficient management of computing resources and cooling units. A major obstacle in the analysis of data centers is the lack of a holistic simulator, where the impact of new computing resource (or cooling) management techniques can be tested with diffierent designs (i.e., layouts and configurations) of data centers. To fill this gap, this paper proposes Green Data Center Simulator (GDCSim) for studying the energy efficiency of data centers under various data center geometries, workload characteristics, platform power management schemes, and scheduling algorithms. GDCSim is used to iteratively design green data centers. Further, it is validated against established CFD simulators. GDCSim is developed as a part of the BlueTool infrastructure project at Impact Lab.

Energy Efficiency of Thermal-Aware Job Scheduling Algorithms under Various Cooling Models

One proposed technique to reduce energy consumption of data centers is thermal-aware job scheduling, i.e. job scheduling that relies on predictive thermal models to select among possible job schedules to minimize its energy needs. This paper investigates, using a more realistic linear cooling model, the energy savings of previously proposed thermal-aware job scheduling algorithms, which assume a less realistic model of constant cooling. The results show that the energy savings achieved are greater than the savings previously predicted. The contributions of this paper include: i) linear cooling models should be used in analysis and algorithm design, and ii) although the job scheduler must control the cooling equipment to realize most of the thermal-aware job schedule’s savings, some savings can be still achieved without that control.

Energy Proportionality and the Future: Metrics and Directions

This paper proposes a pair of quantitative metrics, namely the idle-to-power ratio (IPR) and the linear deviation ratio (LDR), to be used together for measuring the energy proportionality of computing systems. The metrics are applied to the publicly available SPECPower_ssj2008 benchmark results, yielding a partly grim trend of energy proportionality, which has to be addressed and possibly rectified by the computer industry. The paper also proposes future research directions on studying the effects of energy proportionality on software-based energy-saving techniques.

Dynamically adapting registration areas to user mobility and call patterns for efficient location management in PCS networks

In this paper, we propose an extension to the personal communication services (PCS) location management protocol which uses dynamically overlapped registration areas. The scheme is based on monitoring the aggregate mobility and call pattern of the users during each reconfiguration period and adapting to the mobility and call patterns by either expanding or shrinking registration areas at the end of each reconfiguration period. We analytically characterize the trade-off resulting from the inclusion or exclusion of a cell in a registration area in terms of expected change in aggregate database access cost and signaling overhead. This characterization is used to guide the registration area adaption in a manner in which the signaling and database access load on any given location register (LR) does not exceed a specified limit. Our simulation results show that it is useful to dynamically adapt the registration areas to the aggregate mobility and call patterns of the mobile units when the mobility pattern exhibits locality. For such mobility and call patterns, the proposed scheme can greatly reduce the average signaling and database access load on LRs. Further, the cost of adapting the registration areas is shown to be low in terms of memory and communication requirements.