10.4 Energy Aware Data Center: Design and Management

Printer-friendly version PDF version

Date: Thursday 12 March 2015
Time: 11:00 - 12:30
Location / Room: Chartreuse

Andrea Bartolini, Università di Bologna, IT / ETH Zürich, CH

Andreas Burg, École Polytechnique Fédérale de Lausanne (EPFL), CH

The session covers various topics in improving data center energy efficiency, from hardware acceleration, scheduling to cooling.

TimeLabelPresentation Title
Eunhyeok Park1, Junwhan Ahn2, Sungpack Hong3, Sungjoo Yoo1 and Sunggu Lee1
1POSTECH, KR; 2SNU, KR; 3Oracle, US
Energy efficiency in big data processing is one of key issues in servers. Big data processing, e.g., graph computation and MapReduce, is characterized by massive parallelism in computation and a large amount of fine-grained random memory accesses often with structural localities due to graph-like data dependency. Recently, GPU is gaining more and more attention for servers due to its capability of parallel computation. However, the current GPU architecture is not well suited to big data workload due to the limited capability of handling a large number of memory requests. In this paper, we present a special function unit, called memory fast-forward (MFF) unit, to address this problem. Our proposed MFF unit provides two key functions. First, it supports pointer chasing which enables computation threads to issue as many memory requests as possible to increase the potential of coalescing memory requests. Second, it coalesces memory requests bound for the same cache block, often due to structural locality, thereby reducing memory traffics. Both pointer chasing and memory request coalescing contribute to reducing memory stall time as well as improving the real utilization of memory bandwidth, by removing duplicate memory traffics, thereby improving performance and energy efficiency. Our experiments with four graph computation algorithms and real graphs show that the proposed MFF unit can improve the energy efficiency of GPU in graph computation by average 54.6% at a negligible area cost.

Download Paper (PDF; Only available from the DATE venue WiFi)
Shuo Liu1, Soamar Homsi1, Ming Fan1, Shaolei Ren1, Gang Quan1 and Shangping Ren2
1Florida International University, US; 2Illinois Institute of Technology, US
Data centers have been widely employed to offer reliable and agile on-demand web services. However, the dramatic increase of the operational cost, largely due to the power consumptions, has posed a significant challenge to the service providers as services expand in both scale and scope. In this paper, we study the problem of how to improve resource utilization and minimize power consumption in a data center with guaranteed quality-of-service (QoS). Different from a common approach that separates requests with different QoS levels on different servers, we devise an approach to pack requests of the same service --- even with different QoS requirements --- into the same server to improve resource usage. We also develop a novel method to improve the system utilization without compromising the QoS levels by removing potential failure requests. Experimental results show superiority of our approach over other widely applied approaches.

Download Paper (PDF; Only available from the DATE venue WiFi)
Christian Conficoni1, Andrea Bartolini2, Andrea Tilli1, Gianpietro Tecchiolli3 and Luca Benini4
1Università di Bologna, IT; 2Università di Bologna, IT / ETH Zürich, CH; 3Eurotech, IT; 4Università di Bologna / ETH Zürich, IT
Hot-water liquid cooling is a key technology in future green supercomputers as it maximizes the cooling efficiency and energy reuse. However the cooling system still is responsible for a significant percentage of modern HPC power consumption. Standard design of liquid-cooling control relies on rules based on worst-case scenarios, or on CFD simulation of portion of the entire system, which cannot account for all the real supercomputer working conditions (workload and ambient temperature). In this work we first introduce an analytical model, based on lumped parameters, which can effectively describe the cooling components and dynamics, and can be used for analysis and control purposes. We then use it to design an energy-optimal control strategy which is capable to minimize the pump and chiller power consumption while, meeting the supercomputer cooling requirements. We validate the method with simulation tests, taking data from a real HPC cooling mechanism, and comparing the results with state-of-the-art commercial cooling system control strategies.

Download Paper (PDF; Only available from the DATE venue WiFi)
Karim Kanoun1 and Mihaela van der Schaar2
1École Polytechnique Fédérale de Lausanne (EPFL), CH; 2University of California, Los Angeles, US
Several techniques have been proposed to adapt Big-Data streaming applications to resource constraints. These techniques are mostly implemented at the application layer and make simplistic assumptions about the system resources and they are often agnostic to the system capabilities. Moreover, they often assume that the data streams characteristics and their processing needs are stationary, which is not true in practice. In fact, data streams are highly dynamic and may also experience concept drift, thereby requiring continuous online adaptation of the throughput and quality to each processing task. Hence, existing solutions for Big-Data streaming applications are often too conservative or too aggressive. To address these limitations, we propose an online energy-efficient scheduler which maximizes the QoS (i.e., throughput and output quality) of Big-Data streaming applications under energy and resources constraints. Our scheduler uses online adaptive reinforcement learning techniques and requires no offline information. Moreover, our scheduler is able to detect concept drifts and to smoothly adapt the scheduling strategy. Our experiments realized on a chain of tasks modeling real-life streaming application demonstrate that our scheduler is able to learn the scheduling policy and to adapt it such that it maximizes the targeted QoS given energy constraint as the Big-Data characteristics are dynamically changing.

Download Paper (PDF; Only available from the DATE venue WiFi)
12:30End of session
Lunch Break, Keynote lecture from 1320 - 1350 (Room Oisans) in Les Écrins

Coffee Break in Exhibition Area

On all conference days (Tuesday to Thursday), coffee and tea will be served during the coffee breaks at the below-mentioned times in the exhibition area.

Lunch Break

On Tuesday and Wednesday, lunch boxes will be served in front of the session room Salle Oisans and in the exhibition area for fully registered delegates (a voucher will be given upon registration on-site). On Thursday, lunch will be served in Room Les Ecrins (for fully registered conference delegates only).

Tuesday, March 10, 2015

Coffee Break 10:30 - 11:30

Lunch Break 13:00 - 14:30; Keynote session from 13:20 - 14:20 (Room Oisans) sponsored by Mentor Graphics

Coffee Break 16:00 - 17:00

Wednesday, March 11, 2015

Coffee Break 10:00 - 11:00

Lunch Break 12:30 - 14:30, Keynote lectures from 12:50 - 14:20 (Room Oisans)

Coffee Break 16:00 - 17:00

Thursday, March 12, 2015

Coffee Break 10:00 - 11:00

Lunch Break 12:30 - 14:00, Keynote lecture from 13:20 - 13:50

Coffee Break 15:30 - 16:00