UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN SABBATICAL LEAVE REPORT SUBMISSION COVER SHEET The Sabbatical Leave Policy states: "Upon returning from sabbatical leave, the faculty member must submit a written report, in triplicate, on his or her study and travel during the leave. Submission should be made through the normal channels (executive officer, dean/director, Provost). In addition to describing activities undertaken during the leave, the report should address how the sabbatical leave enhanced the individual's value to the University and how it improved the applicant's capabilities for carrying out the objectives of the unit and the University." Sabbatical leave reports are due to the unit executive officer within 60 service days after the start of the next academic term following the return from the sabbatical leave. TO BE COMPLETED BY FACULTY MEMBER Name: __LAXMIKANT KALE______________________ Phone: ___(217) 244-0094_________________ Dept./Unit: __Computer Science________________ Address: _1340 W. Springfield Ave._____________ Leave Taken (year, semester): __Fall 2001 – Spring 2001_________________________________________ This sabbatical leave of absence was taken in accordance with the policies and guidelines stated in communication No. 19 of the Provost's Communications. Faculty: Laxmikant Kale Date: 11/13/2002 REPORT TRANSMITTED VIA: Dept Head: Dean: Date: Date: Provost: Date: Please attach your sabbatical leave report in triplicate to this transmittal cover sheet. Sabbatical Report Laxmikant Kale 8/21/2001 – 5/20/2002 My plans during the sabbatical was to study the issues involved in programming extremely large PetaFLOPS class parallel machines that are expected to emerge in the next few years. One of the specific design to study was IBM’s Blue Gene machine. The timing I chose for this topic turned out to be quite right. During the last year PetaFLOPs computing received a substantial boost, because of the unveiling of the Japanese parallel supercomputer called the Earth Simulator (a 40 TeraFLOPS machine) , IBMs announcement of concrete plans to build a different design of a 360 TF machine, a large DARPA initiative which awarded five vendors contracts to design next generation PetaFLOPS class machines, and DOE’s initiatives to build and use other parallel machines. I received a grant for developing programming environments for such machines at the beginning of the academic year. I established a collaborative relationship with IBM’s research team leading the design of the new machines (called BG/L), formerly led by Marc Snir, now led by Manish Gupta and Jose Moreira. Several interactions also took place with Frederica Darema of NSF who is now an active supporter of PetaFLOPS research. In early Spring, I took part in a small but influential workshop on PetaFLOPS computers, organized by Thomas Sterling and others. The upcoming DOD initiatives on these topics were dicussed there. I think that I was able to articulate a viewpoint about how these systems should be programmed, and how the design of the machines should be influenced by this programming model, and ensured an intellectual space for this viewpoint in the national agenda. This was possible because of the focus I was afforded by the sabbatical during the Fall semester. In March, I participated in an NSF organized workshop on Performance Engineering and PetaFLOPS systems organized by Dr. Darema. This led to a meeting at IBM in May for a small group of researchers focused on PetaFLOPS machine and software design. I was invited to a Blue Gene workshop at Lake Tahoe in the summer, where my approach received a wide audience, and support. Research connections established during this time have continued to develop since then. Department of Energy has emerged as a major agency for the future of parallel computing. As a result of my sabbatical year efforts, I was invited to a PetaFLOPS workshop organized at Sandia National Labs in June 2002. I am leading and participating in a few subgroups that were formed there, including one on rethinking of memory hierarchies for future machines. This workshop was followed one of new operating systems for future machines to which I was invited. As a result of my effort during the year, I have been invited to join the efforts on developing parallel programming environments led by Rusty Lusk of Argonne National Laboratory. I was invited to present a state of the field talk at the annual LACSI conference, a DOE organized conference focusing parallel computing and its applications. Also, SUN Microsystems, one of the vendors contracted by DARPA for PetaFLOPS machine design, has expressed interest including me as a research partner in their project. My research administration burden increased significantly during this year, since I received 3 major grants from NSF: The PataFLOPS grant, an ITR grant focused on large scale parallelization of Quantum Mechanical (ab Initio) calculations of electronic structures, and one on Material Simulation (involving collaborations with several Univ. of Illinois faculty, including Profs. Duane Johnson, Robert Haber and Jonathan Dantzig). Further, two large grants were renewed during this year: The NIH grant, with Profs. Schulten and Skeelas the lead PI (about 1.5 M/year) was renewed for five years in the spring. The DOE grant to the Rocket Center, head by Prof. Heath with 20+ faculty collaborators was also renewed during this year. I believe that my involvement with PetaFLOPS research contributed to the quality of the proposals and site visit presentations. Initiating the research on the new grants required a substantial amount of effort, including hiring and training of students and planning of research. I continued to guide the students in my research group actively during the year, as a result. One substantial achievement of the research conducted during the past year was a paper on parallelization of the Molecular Dynamics program NAMD to an unprecendented level of performance on the 3000 processor machine at Pittsburgh Supercomputing center (currently the third fastest machine in the world). This paper is one of the two nominated for the prestigious Gordon Bell award in the “special” category, to be awarded at SC2002 conference. New Research collaborations were begun as a result of my efforts to seek applications of PetaFLOPs computers. One of these led to a five year ITR grant from NSF to parallelize Computational Cosmplogy and Astronomy applications in collaboration with Prof. Thomas Quinn of Univ. of Washington. The knowledge I developed and I acquired during the sabbatical year on PetaFLOPS systems is invaluable in my ability to teach courses relevant to parallel computing, including CS320 that I wil be teaching in the Spring 2002, and CS433 which I often teach. I will also be presenting a tutorial on performance optimizations on large scale parallel machines at SC2002. I believe that my ability to present such a tutorial as well as its acceptance by the community and organizers of SC2002, was significantly enhanced by my sabbatical year studies. In summary, the focused effort on PetaFLOPS systems that I was able to make due to the sabbatical leave has led to a substantial increase in my knowledge of such systems, enhanced my ability to teach CS courses, increased my influence in the research and development community in this area, broadened my ideas as well as collaborations, and increasing funding opportunities for my research.