Speaker: Anthony Rowstron (Principal Researcher, Microsoft Research Cambridge)



Title: Converged Data Centers



Abstract:

We have been exploring what happens when you combine ideas from distributed systems, networking, high-performance computing (HPC) and recent advances in hardware and apply them to commodity data center clusters. The underlying motivation is that building distributed systems is hard, and some of that complexity comes from having to handle problems that are consequences of design choices of the underlying (hardware) platform. For example, trying to reverse engineer properties of the underlying network topology, in order to ensure locality in a distributed system. Running distributed systems across the Internet it is hard to change the underlying platform. In a data center there is a single entity that owns the servers, networking and controls the software that runs on the servers, and hence it is very feasible to change the base design. We have been building and exploring a number of different clusters with different properties from the clusters traditional used in data centers. These are combined with software stacks designed to ease the development of distributed applications, making them higher performance or easier to develop. The talk will use two motivating examples to demonstrate the concepts. The first example is the based on the CamCube project which explores using different interconnects, inspired by the HPC world, to run distributed applications like Map Reduce. The second example is looking at how hardware trends should be changing the way we think of implementing some services in the data center. The general theme is that we are closing the gap between hardware and software, leading to converged data centers.

Biography:

Antony Rowstron is a Principal Researcher at Microsoft Research in Cambridge, UK and leads the Systems and Networking Group. He currently works in Networked Systems, or in other words at the intersection of Networking, Systems and Distributed Systems. He is best known for his work on structured overlays or DHTs. He is currently interested in understanding how to build future generation data center clusters and in understanding how to make cloud data centers more predictable for the tenets that use them. He has served on the program committees for many major conferences, covering the areas of networked systems, networking and systems. He received an MEng degree in Computer Systems and Software Engineering in 1993 from the University of York, UK, and a DPhil degree in Computer Science in 1996 from the University of York, UK. In 1996 he moved to the Computer Laboratory at Cambridge University, UK as a Research Associate and then moved to the Laboratory for Communications Engineering in the Engineering Department, Cambridge University, UK as a Senior Research Associate. In mid 1999 he moved to Microsoft Research Ltd in Cambridge, UK where he has been since.