Gustavo Alonso gave the keynote talk ”Accelerating Data Science“ at the Fourth International Workshop on In-Memory Data Management and Analytics (IMDM 2016) collocated with VLDB 2016 in New Delhi.
Data science or big data, whatever one wants to call it, raises important challenges in terms efficient processing. One the one hand, the application demands are becoming more stringent (more data, more complex analysis, faster results, larger workloads, etc.). On the other hand, hardware and computing platforms are in a complex phase with little stability in terms of architectures and lacking an overall direction. In this talk I will discuss the problem, arguing that there is an opportunity for specialized designs departing from general purpose systems. I will illustrate the point with examples from our research and then show how we are exploiting reconfigurable hardware (FPGAs) to explore a wide range of architectural designs, new algorithms for data processing, and redesigning the entire system stack to better support data science. The talk will conclude with a number of ideas on how the database community can contribute to the development of new hardware and how to orchestrate a more coherent, collective research agenda.