Modern cloud services operate at significant and increasing scale. The growth of these services has led to the need for automated management in the form of orchestrators.
The significant uptake of orchestrators has led to their expansion out of private datacenters, into the public cloud, and now even towards the edge of the network. These are environments for which they are not designed.
In this talk, I'll examine orchestrator design, focusing on the global state they maintain in their central datastores. I'll give an overview of the orchestration problem and describe a lightweight formalisation using model checking, applied to a popular orchestration platform. With this model we can explain observed failures arising from changes in the consistency model.
Using insights from this model I then describe two new datastores to support the control-plane of orchestration platforms, for the public cloud and the near-edge.
Bio: Andrew is a final year PhD student, having just completed his viva. He has worked in the SRG throughout his time and at Microsoft Research. He focuses on the implementation of systems and ways to improve their reliability.
