Industry research has a rich legacy in computer science [9]. However, as opposed to the blue-sky approach to research, increasingly there is a trend to align industry research more closely with the products. This is manifested in several new trends in industry research: (i) emphasis on product impact, e.g., improving existing products or seeing new ones coming around the bend, (ii) popularity of blended job functions, such as scientist, research scientist, and data scientist, and (iii) setting up research teams that are integrated within the product organization to forge closer collaborations. The latter is a case in point in Azure Data group at Microsoft, where the Gray Systems Lab (GSL) [2] is an applied research team within the product group. Such integrated research labs offer beautiful opportunities for combining research with product impact. Yet, due to their product focus from the get-go, applied research labs could also be challenging to get started. Fortunately, it turns out that there are a set of things that new researchers could do in order to set themselves up for success in a product group.In this paper, we describe the key lessons learned from the CloudViews project [1] at GSL. CloudViews project started with identifying and reusing common subexpressions in big data workloads at Microsoft, however, it was also successful in spinning up a number of followup projects, establishing the GSL ties with the SCOPE team, and seeding the bigger vision of workload optimization, resulting in the Peregrine [7] and Flock [4] projects. Although the lessons we discuss below are derived from the CloudViews project at GSL, we believe the learnings are applicable to other industry research settings as well. Note that there could be several successful ways of going about applied research, however, in this paper, we only discuss the things that we found useful in our experience from the CloudViews project.
展开▼