While the transistor density is continuing to grow exponentially according to Moore’s Law, it is no longer improving processor speeds. Instead, chip manufacturers are packaging an increasing number of cores to meet the new challenges in power consumption and heat dissipation. We entered in a new computing era when energy efficiency becomes an economic necessity and a serious problem for data centres because of the growing power consumption and the increased cost of electricity. Often, the initial investment for purchasing a computing resource is surpassed by the electricity costs for its operation.
Reducing power consumption has thus become a key issue for the industry and research fields alike, not only because of these pressing economical issues, but also for environmental and marketing reasons.
Our research in the field of energy efficiency is centred around the following objectives:
- Energy instrumentation: design and development of automated, live energy instrumentation methods for server-grade hardware, using high precision, professional devices;
- Energy analysis and modelling: characterizing the behavior of homogeneous and heterogeneous hardware in terms of energy under various computational loads
- Energy efficient resource management: provisioning resources that enforce energy-aware QoS parameters
- Multi-objective optimization:
- optimizing the scheduling of distributed workloads in federated data centres (e.g. Cloud) with multiple concurrent objectives; we optimize for energy, cost and reliability, along with the traditional performance objective.
- optimizing code regions of parallel applications in HPC environments with multiple concurrent objectives such as time, resource usage and energy. We perform both both source code transformations and tuneable run-time parameters using the Insieme compiler framework.
Projects
GEMSCLAIM – GreenEr Mobile Systems by Cross LAyer Integrated energy Management
Personal computing currently faces a rapid trend from desktop machines towards mobile services, accessed via tablets, smartphones and similar terminal devices. With respect to computing power, today’s handheld devices are similar to Cray-2 supercomputers from the 1980s. Due to higher computational load (e.g. via multimedia apps) and the variety of radio interfaces (such as WiFi, 3G, and LTE), modern terminals are getting increasingly energy hungry. For instance, a single UMTS upload or a video recording process on today’s smartphones may consume as much as 1.5 Watts, i.e. roughly 50% of the maximal device power. In the near future, higher data rates and traffic, advanced media codecs, and graphics applications will ask for even more energy than the battery can deliver. At the same time, the power density limit might lead to a significant share of “Dark Silicon” at 22nm CMOS and below. Obviously, disruptive energy optimizations are required that go well beyond traditional technologies like DVFS (dynamic voltage and frequency scaling) and power-down of temporarily unused components. The GEMSCLAIM project aims at introducing novel approaches for reducing this “greed for energy”, thereby improving the user experience and enabling new opportunities for mobile computing.
We focus on three novel approaches:
- cross layer energy optimization, ranging from the compiler over the operating system down to the target HW platform
- efficient programming support for energy-optimized heterogeneous Multicore platforms based on energy-aware service level agreements (SLAs) and energy-sensitive tunable parameters
- introducing energy awareness into Virtual Platforms for the purpose of dynamically customizing the HW architecture for energy optimization and online energy monitoring and accounting.
GEMSCLAIM will provide new methodologies and tools in these domains and will quantify the potential energy savings via benchmarks and a HW platform prototype.
EN-ACT – ENergy Aware CompuTing
EN-ACT aims to drastically reduce energy consumption and CO2 emissions of applications and systems based on Information and Communication Technology (ICT). The main goal of the project is to define specifications for the design of energy-aware software (that specifically concerns about the energy consumption, i.e., “green software”). The project will be characterized in terms of energy performance at application level, based on the implementation of platform-independent metrics.
Particular Objectives:
- To develop a framework to identify and monitor energy-relevant parts of the code.
- To define metrics for evaluating performance at the system level, from the application to the hardware.
- To achieve an energy-aware environment consists of a set of specifications, a compiler for the generation of executable applications, and an operating system.
Staff
- Prof. Dr. Thomas Fahringer
- Philipp Gschwandtner PhD
- Dr. Peter Thoman
- Peter Zangerl MSc
- Alexander Hirsch
Former Members
- Assoc. Prof. Dr. Radu Prodan
- Dr. Juan J. Durillo
- Dr. Simon Ostermann
- Ferdinando Alessi, MSc
- Vincenzo De Maio, PhD
- Ivan Grasso, PhD
- Klaus Kofler, PhD
Selected Publications
- Benedict, Shajulin and Gschwandtner, Philipp and Fahringer, Thomas, TOEP: Threshold Oriented Energy Prediction Mechanism for MPI-OpenMP Hybrid Applications. In Proceedings of the Eleventh International Conference on Contemporary Computing (IC3) 2018, Noida, India, 2018
- Durillo, Juan J. and Gschwandtner, Philipp and Kofler, Klaus and Fahringer, Thomas, Multi-Objective Region-Aware Optimization of Parallel Programs. In Parallel Computing, accepted, Elsevier, 2018
- Gschwandtner, Philipp and Hirsch, Alex and Benedict, Shajulin and Fahringer, Thomas, Towards Automatic Compiler-assisted Performance and Energy Modeling for Message-Passing Parallel Programs. In Proceedings of the 13th Workshop on Parallel Systems and Algorithms (PASA) 2018, accepted, Braunschweig, Germany, VDE, 2018
- Kofler, Klaus and Durillo, Juan J. and Gschwandtner, Philipp and Fahringer, Thomas, A Region-Aware Multi-Objective Auto-Tuner for Parallel Programs. In Proceedings of the Tenth International Workshop on Parallel Programming Models and Systems Software for High-End Computing (P2S2) 2017, accepted, Bristol, UK, IEEE Computer Society Press, 2017
- Benedict, Shajulin and R.S., Rejitha and Gschwandtner, Philipp, and Prodan, Radu and Fahringer, Thomas, Energy Prediction of OpenMP Applications using Random Forest Modeling Approach. In Proceedings of the Tenth International Workshop on Automatic Performance Tuning (iWAPT 2015), accepted, Hyderabad, India, IEEE Computer Society Press, 2015
- De Maio, Vincenzo and Prodan, Radu, Evaluating Energy Efficiency of Gigabit Ethernet and Infiniband Software Stacks in Data Centres, Proceedings of the 7th IEEE/ACM International Conference on Utility and Cloud Computing (UCC 2014) (accepted), 2014, IEEE Computer Society
- Gschwandtner, Philipp and Chalios, Charalampos and Nikolopoulos, Dimitrios S. and Vandierendonck, Hans and Fahringer, Thomas, On the potential of significance-driven execution for energy-aware HPC, Computer Science – Research and Development, 2014, p. 1-10
- Gschwandtner, Philipp and Durillo, Juan J. and Fahringer, Thomas, Multi-Objective Auto-Tuning with Insieme: Optimization and Trade-Off Analysis for Time, Energy and Resource Usage, Euro-Par 2014 Parallel Processing, 2014, p. 87-98
- Gschwandtner, Philipp and Knobloch, Michael and Mohr, Bernd and Pleiter, Dirk and Fahringer, Thomas, Modeling CPU Energy Consumption of HPC Applications on the IBM POWER7, Parallel, Distributed and Network-Based Processing (PDP), 2014 22nd Euromicro International Conference on, 2014, p. 536-543
- Durillo, Juan J. and Nae, Vlad and Prodan, Radu, Multi-Objective Workflow Scheduling: An Analysis of the Energy Efficiency and Makespan Tradeoff, Cluster Cloud and Grid Computing (CCGrid), 2013 13th IEEE/ACM International Symposium on, 2013, p. 203-210, IEEE Computer Society
- Durillo, Juan J. and Fard, Hamid Mohammadi and Prodan, Radu, MOHEFT: A multi-objective list-based method for workflow scheduling, Cloud Computing Technology and Science (CloudCom), 2012 IEEE 4th International Conference on, 2012, p. 185-192
- Gschwandtner, Philipp and Fahringer, Thomas and Prodan, Radu, Performance Analysis and Benchmarking of the Intel SCC, Cluster Computing (CLUSTER), 2011 IEEE International Conference on, 2011, p. 139-149
- Thoman, Peter and Moritsch, Hans and Fahringer, Thomas, Topology-Aware OpenMP Process Scheduling, Beyond Loop Level Parallelism in OpenMP: Accelerators, Tasking and More, Lecture Notes in Computer Science, 2010, vol. 6132, p. 96-108, Springer Berlin Heidelberg