Harnessing the power of next-generation shared memory multiprocessors

点击翻译按钮获取中文摘要

After decades of research, distributed computing continues to provide stimulating challenges and opportunities driven by disruptive innovations in hardware design. Breaking the mold of traditional distributed computing models, tomorrow's distributed systems will see large numbers of cores connected in complex ways, new instruction set features, advanced power management capabilities, and non-volatile memories. These design features promise to make future distributed systems more capable than ever before, at the expense of more complex software. The long-term goal of this research program is to harness the power of next-generation distributed computing platforms by revisiting fundamental questions, such as how parallel threads of execution can share resources effectively, and how large systems can survive the failure of software and hardware components. In the short-term, the program will focus on tools and techniques for implementing highly-parallel and fault-tolerant data structures. The specific objectives of the program are: (1) to understand the cost of synchronization in emerging multi-core platforms, and use the insights gained to design efficient shared data structures for such platforms; (2) to develop more robust and efficient mechanisms for distributing data structures across clusters of servers; and (3) to devise improved techniques for protecting shared data structures against failures using non-volatile memory. By embracing new opportunities in distributed computing, this program will advance fundamental research necessary for building novel information systems on tomorrow's hardware platforms, and train highly qualified personnel in skills essential to Canada's technological leadership.