​The scale and importance of data analytics is growing faster than ever, but scalability of both memory capacity and general-purpose processing are not expected to keep up. In this talk, I describe our ongoing efforts to continue scaling the performance and cost of complex analytics via new systems involving technology expected to continue scaling at this time: secondary storage and reconfigurable acceleration (FPGAs). Real-world analytics workloads involve complex and irregular data accesses and processing, which make straightforward mapping to storage and accelerators challenging. We are working to overcome many such hurdles, including storage access latency, page-level access granularity, and accelerator programmability, using a combination of effective abstractions and hardware-software co-optimization. We work with real-world genomics applications as the driving application and show the potential for cheaper, more power-efficient architectures for complex applications.​

​Sang-Woo Jun is an assistant professor in the computer science department of the University of California, Irvine. He has earned his PhD in 2018 at MIT, under the supervision of Professor Arvind. His research focus is on novel, unconventional computer architectures for affordable and scalable data analytics. Some of his projects were fortunate to be recognized by the community, including a ISCA@50 25-year retrospective selection, and a best paper award at FPL.​