Our paper on Firmament has been accepted for publication at OSDI 2016. The paper shows that Firmament's scheduling logic scales to very large clusters of 12,000+ machines despite being fully centralized, and that Firmament makes substantially better placement decisions than other schedulers.
Together with firmament.io, we're starting a new blog series on cluster scheduling. Our first post, on "the evolution of cluster scheduler architectures", discusses how cluster schedulers have changed over the last few years, and where increasingly widely-used open-source orchestration frameworks fit in.
Frank McSherry visited CamSaS and we wrote a blog post, "The impact of fast networks on graph analytics", together. The post shows that graph algorithms such as PageRank can benefit hugely from 10G networks, and demonstrates a 20x speedup over a GraphX/Spark stack.
Our first blog post, "QJump, or: how to be un-British in the data centre" discusses why the guaranteed latency communication offered by QJump is exciting, and what new applications it may enable.