Associate Laboratory Director for Computing, Environment and Life Sciences at Argonne National Lab
The Challenges of Designing and Building a Next-Generation High-Performance Software Library for Genomic Sequence Analysis
In this talk I will discuss a new community project aimed at developing a portable, high-performance library for writing a variety of genomic sequence analysis codes. The library is aimed at current and future parallel architectures, multicore and many core systems as well as accelerator based nodes. The goal is to develop a library of common low-level sequence analysis kernels to serve as a foundation for a wide variety of higher-level tools such as multiple sequence aligners, genome assemblers, short-read mappers, clustering and searching tools, profiling and non-alignment similarity tools, and phylogenetic tools. The effort has the following objectives:
I will motivate the this effort by providing a set of core workflows from the DOE-supported systems biology knowledge base project.
University of Chicago and Argonne National Laboratory
Can We Systematically Evaluate and Exploit Heterogeneous Accelerators? A 10x10 Perspective
In the waning generations of Moore's law, the race is full on to exploit any and all techniques to deliver performance and energy efficiency gains. These include extreme parallelism, extreme voltage scaling (NTV), better-than-worst-case design, and heterogeneous architecture. Of these, perhaps heterogeneity is the most difficult to evaluate for its practical progress and long-term impact. We will discuss a range of efforts, all part of the 10x10 project, to develop the capability to systematically evaluate and exploit heterogeneity. Specifically, these include methods to evaluate comparatively the performance of a range of heterogeneous accelerators. Also, we will discuss results from efforts to assess the broad suitability of both Department of Energy workloads and general-purpose computing workloads for heterogeneous accelerators, and specific opportunities in those areas. Finally, we'll discuss future directions in developing 10x10 architectures and programming systems.