Traditional database operators such as joins are relevant not only in the
context of database engines but also as a building block in many computational and
machine learning algorithms. In this talk, we explore the implementation of distributed
join algorithms in systems with several thousand cores connected by a high-throughput,
low-latency network. We compare radix hash join to sort-merge join algorithms, discuss
their implementation at this scale, show the impact and advantages of RDMA, and
discuss the importance of network scheduling. The experimental results show that the
algorithms we present scale well with the number of cores. At the end of the talk, I will
highlight future research directions (e.g., In-Network Processing) we are currently exploring
at ETH Zurich.