Building Energy-efficient Clusters for I/O-intensive workloads:
A Fast Array of Wimpy Nodes (FAWN).

Prof. Dave Andersen
CMU School of Computer Science

Abstract: Power and cooling have rapidly become a barrier to scaling
computer systems, from individual chips to entire data-centers.  In
this talk, I will present our preliminary work on designing a new
cluster architecture that uses up to an order of magnitude less power
to serve I/O-bound workloads.  Particularly for seek-intensive loads,
today's servers suffer an enormous gap between the capabilities of the
processor and what the storage subsystem can deliver to that
processor, but still spend tens to hundreds of watts powering the CPU
and chipset.  Our new architecture, termed a FAWN--a Fast Array of
Wimpy Nodes--operates by coupling low-power embedded-system-class
processors with fast flash memory, effectively re-balancing the I/O
and computational capabilities of the cluster.  Programming such a
cluster can be challenging; we present our initial work that shows how
a FAWN can be organized to easily support high-volume key-value lookup
workloads, such as those experienced by Facebook or Amazon, as well as
throughput intensive workloads derived from machine learning projects
that need to "grep the web."

Bio: David Andersen is an assistant professor in the Computer Science
department at Carnegie Mellon University.  He received his Ph.D. and
M.S. degrees from MIT, and and received B.S. degrees in Computer
Science and Biology from the University of Utah.  Before joining MIT,
he was a co-founder and CTO of an Internet Service Provider in Salt
Lake City.  His research interests center on computer systems in the
networked environment.