11:45 FPGAs in Database Research – J. Teubner (ETHZ)
FPGAs are programmable chip devices that promise advantages in terms of
performance increase and energy savings. Here I use the
programmablility of FPGAs to model different types of parallelism. My
context is the database domain, and I will show how proper algorithmic
design can guarantee scale-out even to the high degrees of parallelism
that FPGA hardware can offer.
The lessons that we learn in the FPGA world turn out to be valuable also
for the world of modern multi-core architectures. I will show this by
example of a database stream join operator that will scale linearly to
(at least) 30 CPU cores.
The work I present came out of the “Avalanche” project at ETH and was
done jointly with Gustavo Alonso, Rene Mueller, and Louis Woods.