Authors: Justin Y. Shi
The proposed talk will address the fundamental scalability challenge for distributed and parallel applications. Statistic Multiplexing (SM) is a known method for addressing reliability and performance issues for large scale communication systems. When the probability of component failure is greater than zero, SM is the only known method for delivering open-scale communicating systems without scalability limits. This talk focuses on Statistic Multiplexed Computing (SMC). Unlike communication systems, computing applications assume reliable application-level communication based on the statistic multiplexed networking layers. Although these assumptions have helped to simplify application programming tasks, they have also made distributed and parallel application fundamentally not scalable. The fundamental problem is the mismatch between the application-level assumption and the deliverables of the communication layers. The mismatches are clearly displayed in the OSI 7-layer communication model where layers 1-4 (SM layers) are well-protected and 5-7 are without protection. Thus every computing node is a single-point failure in any distributed and parallel application. It seems impossible to eliminate single-point failures unless for some very special purpose applications. This talk will present two SMC architectures: a) compute intensive SMC architecture and b) data intensive SMC architecture. These SMC blue-prints promise to deliver unlimited performance and reliability as we upscale our infrastructure for the two commonly used application types. We will also address the commonly accepted theoretical limitations, such as CAP Theorem, Amdahl’s and Gustafson’s Laws to show that the application scalability “glass ceiling” can indeed be broken.
Justin Y. Shi is also known as Yuan Shi, graduated from Shanghai Jiaotong University in China specialized in Computer Engineering in 1979. He earned his Masters and Ph.D. degrees in Computer Science from the University of Pennsylvania in U.S. in 1983 and 1984 respectively. His dissertation provided a unique solution to the distributed synchronization and termination problems for the Link econometric simulation project led by Professor and Nobel Laureate Lawrence Klein from the Wharton School of University of Pennsylvania.
Dr. Shi’s teaching and research career started at Temple University since 1985. He was granted two patents for the design of large scale heterogeneous parallel computing. He is also the co-inventor of the DBx system with his former student Suntian Song with Parallel Computers Technology Inc. in King of Prussia of Pennsylvania, an independent research and development company he founded since 1997.
Dr. Shi was elected and appointed as the Chairman for the Computer and Information Sciences Department of Temple University from 2007-2009. He has been appointed as the Associate Chairman and Graduate Program Chair since 2009. He is a member of the Graduate Board of Temple University, a board member for Center of Responsible Journalism of Temple University and a member of Technology Advisory Committee for the Pennsylvania State Benjamin Franklin Technology Partnership Program.
Since early 1980’s, Dr. Shi’s research has been focused on programming paradigms and architectures for extreme scale systems using volatile resources. He is the architect and chief programmer for the Synergy parallel processing system. His recent discoveries include the imperfections in OSI 7-layer communication model and statistic multiplexed computing architectures for compute intensive and data intensive applications. His research has been supported by the U.S. National Science Foundation, National Institute of Health and other U.S. Government entities. His 2012 projects include the construction of Temple University private HPC Clout (TCloud), peer-to-peer Hadoop File System (p2pHDFS) and Growshare.net – an auction-based civic exchange mobile social network.