Academic datacentres are nowhere near as large as Google, Amazon or Facebook. Yet a datacentre of 100 racks can deploy many of the same techniques as a datacentre of 10,000 racks, and reap the same benefits. In this talk, we'll explain how we took the lessons from our big friends, and applied them to our datacentre. In a few years, SURFsara (the Dutch national supercomputing centre with 170 racks) migrated from a network consisting of traditional vendor brands like Juniper, Cisco and Arista, with mostly manual NOC operations, to a datacentre with white label network devices, automated tooling, a scalable Clos network with overlay network, and telemetry-based monitoring and metrics collection. In this talk, we describe the steps we took so far at SURFsara, what steps are still ahead of us, and what we think the future holds for networking in medium- or small-scale datacentres. This paper describes our results. While we try to generalise our findings to other networks, your mileage may vary.
Networks · Services · People | Code of Conduct | Privacy Policy
