This project addresses optimizing energy efficiency in the execution of parallel algorithms.
High energy cost is a salient constraint when running large scale parallel applications on the next generation of supercomputers that contain heterogeneous multicore processors and interconnections, motivating a rethinking of conventional approaches to modeling, designing and scheduling parallel tasks by taking energy-efficiency into consideration.
In this collaborative research, this team explores energy-efficient parallel task design, scheduling, and implementation and develops an power profiling tool (PowerPack) that can measure decomposed runtime power consumption of different computing components (e.g. processors, memory, networks and disks) when running large scale parallel applications.
The results of the research will be widely disseminated by maintaining an active project website, publishing peer-reviewed journal and conference papers, making the code available to other researchers, and presenting the research results in professional meetings. The availability of the research outcomes will provide ample opportunities for other researchers to further study the energy-efficiency of parallel applications. Through the collaboration of Texas State University ? San Marcos, Colorado School of Mines, and the Marquette University, PIs promote teaching, learning, and training by exposing graduate and undergraduate students to technological underpinnings in the fields of high performance computing in general and energy-efficient computing in particular. The close partnerships with a number of universities, data centers and national laboratories will also facilitate the broad dissemination of the proposed energy-efficient parallel tasks designing and scheduling techniques as well as the developed power profiling toolkits.
High energy cost has become a salient constraint when running large scale parallel applications on the next generation of supercomputers that contain heterogeneous multicore processors and interconnections. The primary goals of this collaborative CNS grant include: (1) investigating the impact of different parallel task design strategies on performance and energy-efficiency; (2) exploring energy-aware parallel scheduling algorithms for heterogeneous multicore systems; (3) developing an easy-to-use profiling toolkit that can obtain decomposed runtime power consumption characteristics of parallel tasks. Three universities (Marquette University, Texas State University, and University of California - Riverside) are involved in this collaborative grant (#1116691, #1118043, and #1304969). The proposed research goals have been accomplished and the outcomes derived from this collaborative grant are summarized below: 1) Research Activities: The PIs and students have researched energy-efficient algorithms and technologies through eight supported projects, including eTune power analysis framework, energy analysis and optimization for MapReduce data movements, energy consumption analysis of parallel sorting algorithms, energy-efficient scheduling for multicore systems with bounded resources, self-adaptive resource scheduling for heterogeneous cloud systems, energy efficient parallel matrix multiplication for DVFS enabled clusters, energy efficient parallel matrix multiplication via pipeline broadcasting, and intelligent prefetching for reducing energy consumption on large-scale hybrid storage systems. These projects have generated a number of novel algorithms and new studies, which contribute to the green computing discipline. 2) Publications: By the time when this report is submitted, nine peer-reviewed papers have been published in highly recognized IEEE/ACM sponsored conferences/workshops, which include the International Green Computing Conference (IGCC), the IEEE International Conference on Green Computing and Communications (GreenCom), the International Conference on Parallel Processing (ICPP), IEEE Cluster, the ACM Cloud and Autonomic Computing Conference (CAC), and the IEEE International Performance Computing and Communications Conference. 3) Training: At the Marquette University site, four Marquette undergraduate students and five graduate students, and one undergraduate student from University of Wisconsin – Madison participated in this research project. The undergraduate students gained research experience and showed strong research potentials. Ryan Vogt, a junior at the Marquette University in Fall 2013, was a co-author of a conference workshop paper and accepted by the Experiencing HPC for Undergraduates program of the Supercomputing conference (SC13). Patrick Millar, a senior at the Marquette University, was a co-author of a poster submitted to SC13. Brian Hunter from UW Madison is now applying for graduate schools. Graduate students involved in the projects were working to improve in research background, skills, and publication records. Similar training was also conducted at the other two institutes on this collaborative research. 4) Education: The research findings derived from this collaborative grant have been integrated into various levels of classes taught by the PIs at the Marquette University and two other collaborating institutions. Students from these classes are able to leverage the relevant topics to gain first-hand experience and intuition in the fundamental concepts and research frontiers of green computing. 5) Broad Impact: The PIs strive to attract underrepresented students and minority students involved in the research projects supported by this grant. At the Marquette University site, Heather Bort, a female graduate, was accepted by both the Grace Hopper Celebration of Women in Computing and Broader Engagement program of the Supercomputing conference (SC13); William McRae, an African American graduate student was recruited to work on this project; Nurulayuni Abd. Rashid, a female undergradurate was also engaged in this project. Similar efforts were also made in other two institutes. 6) Released Data: The research findings of this project are published in conferences or workshops for public knowledge. The eTune energy profiling system design and software is available to public at http://hpcl.mscs.mu.edu/wiki/index.php/Main_Page#eTune. The energy efficient scheduling code developed by PI Zong has been released to the community at http://cs.txstate.edu/~zz11/software/scheduling/. In summary, PI Ge collaborates closely with two other PIs (Zong and Chen) from Texas State University and University of California - Riverside on this collaborative research project and the PIs have successfully accomplished the proposed research goals. The research projects supported by this grant provide numerous opportunities to undergraduate and graduate students at three institutions to participate and gain valuable experiences in high performance computing research in general and green computing research in particular. The research approaches and results have been introduced into both undergraduate and graduate level courses to benefit a large group of students at three institutions. All published papers have been made publicly accessible through the Internet, contributing to national and international green computing research community.
