News and Announcements
RaPyDLI Underway
From left to right: Judy Qui, Jakub Kurzak, Piotr Luszczek, Gregor von Laszewski, Geoffrey Fox, and Terry Moore.
ICL recently teamed up with Geoffrey Fox from Indiana University and Andrew Ng from Stanford to take on a deep learning project called RaPyDLI (Rapid Python Deep Learning Infrastructure). The NSF-funded project will allow users to program deep learning models in Python, port them to supercomputers, and scale-out to cloud systems, thereby bringing the performance of HPC to deep learning. RaPyDLI will support GPU accelerators and Intel Xeon Phi coprocessors and a broad range of storage approaches including files, NoSQL, HDFS, and databases.
Terry Moore, Jakub Kurzak, and Piotr Luszczek recently met with the rest of the RaPyDLI team to establish an open line of communication and harden the strategy for the RaPyDLI software infrastructure.
ICL now an Intel Parallel Computing Center
The Innovative Computing Laboratory recently became the newest Intel Parallel Computing Center (IPCC). The objective of ICL’s IPCC is the development and optimization of numerical linear algebra libraries and technologies for applications, while tackling current challenges in heterogeneous Intel Xeon Phi coprocessor-based High Performance Computing. In collaboration with Intel’s MKL team, the IPCC at ICL will modernize the popular LAPACK and ScaLAPACK libraries to run efficiently on current and future manycore architectures, and will disseminate the developments through the open source MAGMA MIC library.
New ICL Website Released
As we announced in last month’s newsletter, David Rogers has been hard at work developing ICL’s new web presence. The new ICL website utilizes a Drupal back end and features a completely new look and advanced publications database. Well, as of October 1st, the website is up and running. There are still some new features that will be rolled out in the coming months, but the core site is in place and functional. ICLers are encouraged to check it out!
Conference Reports
EuroMPI/ASIA 2014
On September 9 – 12, ICL’s Aurelien Bouteiller and George Bosilca made their way to Kyoto, Japan for the 21st EuroMPI/Asia meeting. While previous meetings were typically held in Europe, this year’s meeting arrives at a new venue in Japan. Location notwithstanding, EuroMPI is the preeminent meeting for users, developers, and researchers to interact and discuss new developments and applications of message-passing parallel computing, and the Message Passing Interface (MPI) in particular.
While at the meeting, Aurelien and George hosted a tutorial for resilient applications using MPI-level constructs before heading to the MPI Forum in Kobe.
CCDSC 2014
ICL’s Jack Dongarra and former ICLer Bernard Tourancheau (now a Professor at the Université Grenoble Alpes) co-charied, while George Bosilca and Anthony Danalis each gave invited talks. George gave his talk, “Mixed Resilience Solutions,” on the second day, while Anthony gave his talk, “Why PaRSEC is the right runtime for Exascale computing,” on the fourth and final day of the workshop. Overall, the meeting had over 50 attendees and nearly 40 individual talks.
MCSoC-14
IEEE’s 8th International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC-14) was held at the University of Aizu, Aizu-Wakamatsu, Japan, on September 23 – 25. MCSoC provides a forum for leading researchers in academia and industry working in areas of embedded Multicore/Many-core SoCs software, tools, and application design.
Jakub Kurzak and Anthony Danalis both gave invited talks at the symposium. Jakub gave his talk, “BEAST: An Automatic Tuner for Numerical Kernels on Accelerators” on the first day, and later gave another talk, “PULSAR: A Lightweight Runtime for Scientifc Computing,” at the University of Tokyo. On the second day, Anthony gave his talk about, “Dataflow based Task Execution through PaRSEC for High Performance Computing Systems.”
Recent Releases
MAGMA MIC 1.2 Released
MAGMA MIC 1.2 is now available. This release provides implementations for MAGMA’s one-sided (LU, QR, and Cholesky) and two-sided (Hessenberg, bi- and tridiagonal reductions) dense matrix factorizations, as well as linear and eigenproblem solvers for Intel Xeon Phi coprocessors. More information on the approach is given in this presentation.
The MAGMA MIC 1.2 release adds usage and performance improvements.
Visit the MAGMA software page to download the tarball.
Interview
Chunyan Tang
Where are you from, originally?
I was born in Suining, Sichuan Province. Sichuan is in the southwest of China and is famous for its namesake feature, which is like a basin surrounded by mountains. My hometown is in the center of the Sichuan Basin.
Can you summarize your educational background?
I earned my BS in Physics from Beijing Normal University and then obtained my PhD in Condensed Matter Physics in 2006 at the Institute of Physics, Chinese Academy of Sciences. I worked at Brookhaven National Lab as a postdoc for about three years. After that I worked as an assistant project scientist at the University of California, San Diego. I moved to Knoxville with my family in 2011. After my son was born I decided to pursue a different career path. I enrolled in the EECS Master’s program at UTK in the fall of 2013.
Tell us how you first learned about ICL.
My husband, Dong Li, works in HPC at Oak Ridge National Lab. He received help from Jack when he applied for a joint faculty position at UTK. When I looked for a graduate supervisor, Jack was my first choice. Fortunately, ICL also provided me with a graduate research assistant position starting in January 2014. I am so lucky to have this opportunity to join ICL.
What made you want to work for ICL?
Jack Dongarra is a big name in the HPC arena. ICL’s research projects are attractive, and the work experience I gain at ICL will be very helpful in my future job hunt. ICL is also a sizeable lab, so there are ample opportunities to network and collaborate with people.
What are you working on while at ICL?
The first project I worked on involved the Common Communication Interface (CCI). Now I am working on a collaborative project with a team at ORNL. We are porting OpenSHMEM to exascale applications (LAMMPS and S3D), and comparing the performance with an MPI implementation. Our next step is to implement MPI on universal common communication substrate (UCCS). We will apply the combination of MPI and OpenSHMEM to exascale applications to get higher performance.
If you weren’t working at ICL, where would you like to be working and why?
If I was not working at ICL, I might find another lab at UTK to proceed with my project in lieu of a thesis. If I wasn’t enrolled in the EECS’s MS program, I might work as an electron microscopist in a company or a lab. Another possibility is to be a stay at home mom.
What are your interests/hobbies outside work?
I like swimming, hiking, and singing. I also play PingPong and badminton.
Tell us something about yourself that might surprise people.
My right arm was dislocated twice. The first dislocation happened when I was about 7 years old. I was jumpy on the way home after school and then suddenly fell down and hurt my arm. The second dislocation happened when I was 12 and learning to ride a bicycle and suddenly fell down…
