Skip to main content
Add to Calendar
2023-10-17 14:00:00
2023-10-17 15:00:00
America/New_York
Distributed Sparse Computing in Python
This is hybrid meeting. The physical room is 32-D463 (Star). The Zoom registration link is https://mit.zoom.us/meeting/register/tJMtdOqgrTwsGNV_k0Nk6JjLMk-G6GFzTRyk.******************IMPORTANT NOTE ABOUT ONLINE REGISTRATION******************- The registration link for 2023 is the same as the link from 2022.- Please save the Zoom link that you receive after you register. This link will stay the same for all subsequent Fast Code seminars.- Zoom does not recognize a second registration, and will not send out the link a second time. The organizers will not be notified of any second registration.- If you have any problems with registration, please contact lindalynch@csail.mit.edu by 12pm on the day of the seminar, so that we can try to resolve it before the seminar begins.*********************************************************************Abstract: The sparse module of the popular SciPy Python library is widely used across applications in scientific computing, data analysis and machine learning. The standard implementation of SciPy is restricted to a single CPU and cannot take advantage of modern distributed and accelerated computing resources. We introduce Legate Sparse, a system that transparently distributes and accelerates unmodified sparse matrix-based SciPy programs across clusters of CPUs and GPUs, and composes with cuNumeric, a distributed NumPy library. Legate Sparse uses a combination of static and dynamic techniques to efficiently compose independently written sparse and dense array programming libraries, providing a unified Python interface for distributed sparse and dense array computations. We show that Legate Sparse is competitive with single-GPU libraries like CuPy and achieves 65% of the performance of PETSc on up to 1280 CPU cores and 192 GPUs of the Summit supercomputer, while offering the productivity benefits of idiomatic SciPy and NumPy.Bio: Rohan Yadav is a fourth-year computer science Ph.D. student at Stanford University, advised by Alex Aiken and Fredrik Kjolstad. He is generally interested in programming languages and computer systems, with a focus in systems for parallel and distributed computing.
32-D463 (Star)
Add to Calendar
2023-11-21 14:00:00
2023-11-21 15:00:00
America/New_York
High-Performance GPU Code Generation for Mining Motifs in Temporal Graphs
This is hybrid meeting. The physical room is 32-D463 (Star). The Zoom registration link is https://mit.zoom.us/meeting/register/tJMtdOqgrTwsGNV_k0Nk6JjLMk-G6GFzTRyk.******************IMPORTANT NOTE ABOUT ONLINE REGISTRATION******************- The registration link for 2023 is the same as the link from 2022.- Please save the Zoom link that you receive after you register. This link will stay the same for all subsequent Fast Code seminars.- Zoom does not recognize a second registration, and will not send out the link a second time. The organizers will not be notified of any second registration.- If you have any problems with registration, please contact lindalynch@csail.mit.edu by 12pm on the day of the seminar, so that we can try to resolve it before the seminar begins.*********************************************************************Abstract:Temporal motif mining is the task of finding the occurrences of subgraph patterns withina large input temporal graph that obey the specified structural and temporal constraints.Despite its utility in several critical application domains that demand high performance(e.g., detecting fraud in financial transaction graphs), the performance of existing softwareis limited on commercial hard- ware platforms, in that it runs for tens of hours. In this talk,I will present Everest - a system that efficiently maps the workload of mining (supportsboth enumeration and counting) temporal motifs to the highly parallel GPU architecture.Using input temporal graph and a more expressive user-defined temporal motif querydefinition, Everest generates an execution plan and runtime primitives that optimize theworkload execution by exploiting the high compute throughput of a GPU. Everestgenerates motif-specific mining code to reduce long-latency memory accesses andfrequent thread divergence operations. Everest incorporates novel low-cost runtimemechanisms to enable load balancing to improve GPU hardware utilization. To supportlarge graphs that do not fit on GPU memory, Everest also supports multi-GPU executionby intelligently partitioning the edge list that prevents inter-GPU communication. Everesthides the implementation complexity of presented optimizations away from the targetedsystem user for better usability. Our evaluation shows that, using proposed optimizations,Everest improves the performance of a baseline GPU implementation by 19x, on average.Speaker Bio:Nishil Talati is an Assistant Research Scientist (Research Faculty) at the CSE departmentof University of Michigan. He earned his PhD from University of Michigan. Nishil’sresearch interests include computer architecture and systems software design forimproving the performance of modern data-intensive workloads. His research is publishedat several top-tier venues including ISCA, MICRO, HPCA, ASPLOS, and others. Nishil’swork has been recognized as the 2021 HPCA best paper award, 2023 DATE best paperhonorable mention, and 2023 IISWC best paper nominee.Relevant Publication:Y. Yuan, H. Ye, S. Vedula, W. Kaza, and N. Talati, "Everest: GPU-Accelerated System for Mining TemporalMotifs," at International Conference on Very Large Databases (VLDB 2024).
32-D463
