Alex Breslow

Publications

Hailong Yang, Alex Breslow, Jason Mars, and Lingjia Tang. Bubble-Flux: Precise Online QoS Management for Increased Utilization in Warehouse Scale Computers. In the Proceedings of ISCA'13. [pdf|bib]

Alex D. Breslow, Leo Porter, Ananta Tiwari, Michael A. Laurenzano, Laura Carrington, Dean M. Tullsen, and Allan E. Snavely. The Case for Colocation of HPC Workloads. Concurrency and Computation: Practice and Experience; Special issue on the Analysis of Performance and Power for Highly Parallel Systems (CCPE 2013). [pdf]|[bib]

A. Danner, J. Baskin, A. Breslow, and D. Wilikofsky. Hybrid MPI/GPU Interpolation for Grid DEM Construction. In Proc. ACM Symposium on Advances in Geographic Information Systems, pages 299-308, 2012. [pdf|bib]

Research Statement

I am deeply interested in the area of general computing systems, as these systems pose many challenges and because their efficient operation is imperative for a productive society. As more and more computation moves to centralized data centers, the efficiency and throughput of these systems will determine fundamentally what is computable, and thus what we as a society can achieve.

My recent research has attacked the large scale system research question from an High Performance Computing (HPC) perspective. However, I am widening my research scope to address challenges found in the commercial data center realm. My goal is to develop algorithmic and software mechanisms that are general enough to span both scientific and enterprise cloud infrastructures.

In addition to these current endeavors, I am interested in areas relating to hardware-software codesign and the art of elegant parallel programming, as advances in these fields are now the principal mechanism by which we can continue to get significant increases in performance every 18 or so months with Moore's Law. Besides these aspects of the parallel programming challenge, I continue to be fascinated by the design of novel parallel algorithms that allow us to crunch through big data by harnessing the computing power of distributed computing environments.

I currently work with the excellent folks at the Performance Modeling and Characterization Lab at SDSC, and as of January 2013 have begun joint work with Martin Schulz at Lawrence Livermore National Laboratory on projects related to exascale computing. Up until recently, I was advised by the exceptional Allan Snavely; however, on July 14, 2012, he passed away from a heart attack shortly after cycling up Mount Diablo in Northern California. I will always remember his kindly nature and be ever grateful for his unrelenting support of my growth as a researcher.

My work with Allan Snavely identified that there is a substantial performance and energy efficiency benefit to colocating distributed HPC applications from different users across a set of shared servers. This work will appear in February 2013 in the research journal Concurrency and Computation: Practice and Experience. I am currently extending this work and am tackling the two main challenges involved in realizing colocation on supercomputers: establishing fair pricing and deciding which applications should corun. To address the fair pricing challenge, I have developed a light-weight runtime system that prices relative to the performance degradation that an application experiences. This work will be submitted for peer review in early 2013. I aim to address the scheduling challenge shortly thereafter.

Side Interests

Aside from research, I am an avid runner and athlete. I also enjoy playing tennis, rock climbing and practicing yoga. Outside of fitness, I speak Spanish and French with relative fluency and enjoy employing these language skills when traveling in Europe and Latin America.

Acknowledgements

I would like to thank Tia Newhall and Andrew Danner of Swarthmore College for piquing my interest in parallel and distributed computing.

At UCSD, I would be remiss if I did not acknowledge the guidance of the following individuals: Allan Snavely, Ananta Tiwari, Dean Tullsen, Jason Mars, Laura Carrington, Lingjia Tang, Leo Porter and Michael Laurenzano. Each has contributed to my growth as a researcher.

This page layout has been graciously provided by Professor Andrew Danner at Swarthmore College.