Yelick Attends White House Event for National Strategic Computing Initiative
Associate Lab Director Kathy Yelick was among the attendees at a July 29 White House event marking the first anniversary of the National Strategic Computing Initiative (NSCI). Also attending were representatives from all of the participating agencies, leaders from the computing industry, universities and the national labs. According to Yelick, several presentations highlighted new and ongoing efforts in NSCI, as well as the challenges posed by the end of Moore’s Law and some of the opportunities from the convergence of data and high performance computing. Among the topics discussed were the need for partnerships across institutions and workforce development.
CS Summer Students Present Posters Thursday
At 10 a.m. on Thursday, Aug. 4, the Computing Sciences division's summer students will present 40 posters showcasing their work. CS summer students are mentored by CRD, ESnet and NERSC staff. Please stop by the reception held in in Wang Hall (Building 59) room 3101, and support the great work done by our students and their mentors.
Aug. 2 ‘State of the Lab’ Presentation to be Live-Streamed
Director Mike Witherell’s first “State of the Lab” talk will be held from 1 to 2 p.m. tomorrow, Tuesday, August 2, in Building 50 Auditorium. He will discuss Lab-wide initiatives and several other topics, including the ideas led by the Lab at the recent DOE Big Ideas Summit. A live-stream of the talk will be available for those who can’t attend in person.
Cori Phase II is Here!
All 52 cabinets of NERSC's Cori Phase-2 Cray supercomputer arrived last week at Wang Hall. NERSC and Cray staff have been hard at work installing and configuring the system. When it powers up in the coming weeks, Cori will be the largest system for open science based on Intel's Knights Landing processors. In a long-standing NERSC tradition, the system is named in honor of a notable scientist, this time American biochemist Gerty Cori who was the first American woman to win a Nobel Prize in science. (Photo: Glenn Lockwood)
Reminder: 2016 Python Bootcamp August 22 & 23
BIDS is holding its 2016 Python Bootcamp on August 22 and 23 in Evans Hall on the campus of UC Berkeley. The two-day training runs from 8:30 a.m. to 5:00 p.m. each day and is designed to introduce the basics of the Python language to those already familiar with other computing languages (e.g., C, Java, FORTRAN, Lisp). The bootcamp is open to anyone within the UC Berkeley and LBNL community who qualify (that is, who have sufficient programming knowledge). To apply, visit the Python Bootcamp web site and fill out a pre-qualification form. If you are accepted, a small fee to offset the costs will be collected at the time of official registration.
This Week's CS Seminars
Tuesday, August 2
Interdisciplinary Instrumentation Colloquium
Parallel Graph Algorithms Using GraphBLAS + X
10 – 11 a.m., Wang Hall - Bldg. 59, Room 3049
Ariful Azad, Lawrence Berkeley National Laboratory
Combinatorial problems such as graph traversal, graph matching, graph coloring, network flows, shortest paths, minimum spanning trees, and various ordering and selection problems are important components of many scientific applications. Decades of research demonstrated that these combinatorial algorithms are hard to parallelize, especially on distributed-memory systems, because they often have very low computational intensities and employ irregular data accesses and fine-grained communication, which limit their performance. In this talk, I will discuss how these limitations can be partially overcome by designing graph algorithms using standard sparse matrix and vector operations. In particular, I will discuss parallel algorithms for several linear algebraic operations that are part of the GraphBLAS standard (http://graphblas.org), such as sparse matrix-matrix multiplication (SpGEMM) and sparse matrix-sparse vector multiplication (SpMSpV). I will show experimental results of the communication-avoiding algorithms for SpGEMM that scale up to fifty thousands cores of a modern supercomputer, outperforming all the existing implementations at large scale.
I will demonstrate how Graph BLAS primitives accompanied by few non-standard operations can be used to design complex graph algorithms. In particular, I will discuss distributed-memory algorithms for computing a maximum cardinality matching in a bipartite graph and the reverse Cuthill-McKee ordering (RCM). These implementations scale on up to four thousands cores of a Cray XC30 supercomputer (NERSC/Edison), improving over previous state-of- the-art implementations by up to an order of magnitude in terms of scalability and runtime. Finally, I will discuss two data analysis pipelines where these graph algorithms have been used: (a) functional magnetic resonance imaging (MRI) data form human brains and (b) single-cell flow cytometry data from the human immune systems.
I conclude the talk with my view in tackling the theoretical and practical challenges arisen in data analytics due to the explosion of scientific data. I argue that breaking algorithms into algorithmic motifs and developing scalable algorithms for the performance-critical motifs improve both the productivity of application development and the reusability of algorithm libraries. They together create a sustainable software development framework that has a higher chance of adaptation to the rapidly evolving scientific data sets and computing platforms.
Thursday, August 4
Summer Students Poster Session
10 a.m. - 12 noon, Wang Hall (Bldg. 59), Room 3101
Come see what our CS summer students have been working on this summer. This year's poster session will feature 40 different projects from students working with CRD, ESnet and NERSC staff.
Fast Fingerprints for Power System Events
1-2 p.m., Wang Hall (Bldg. 59), Room 4102
David Bindel, Cornell University
In order to operate the power grid aggressively enough to make full use of renewable power, operators need new tools for situational awareness and control. Phasor measurement units (PMUs) have been developed for the past thirty years, but first saw wide-scale production grade deployments in the US after DOE investments funded by the American Recovery and Reinvestment Act of 2009. PMUs report voltage and current phasors thirty or more times every second, promising operators a real-time picture of the state of the grid -- but only with systems and algorithms that transmit the data and analyze this information at similar rates. In this talk, we describe fast analysis using PMU-sensed “fingerprints" of different types of system events (e.g. changes in line status or reconfiguration of substations). Our system, FLiER (Fingerprint Linear Estimation Routine) identifies system changes in close to real time through a novel filtering operation that lets us discard most potential events from consideration with little computation. We describe the elements of our approach, as well as giving an overview of work in progress to improve the quality of our results (and the range of contingencies we can handle) by monitoring the frequency content of transient "ringing" as the system passes from one state to another.