Milton Ngan Email & Phone Number

15Tbps of network capacity, >1Tbps of global backbone and still growing. The growth in content delivery is astounding, and tricky to keep pace with, having only a small team working on it. We are constantly pushing the limits of the hardware available to us, and trying novel approaches to stretch the life of the existing platforms before we switch to the next generation hardware that is only just being released.Last year, I had an opportunity to build a modern datacenter network architecture from the ground up. It was pretty exciting, although it has yet to be used in with full production load. The proof will be coming later this year, when I roll out a smaller scale version of this into our production network. I have spent the 2-3 years working with different big data stacks to aggregate flow data from our global network to give us insight into traffic patterns and DDOS detection. The experience so far has been that for the first 6-8 months things are good while the system gets into a steady state, but then the next few months performance and reliability starts to degrade due to our rapidly increasing traffic volumes. We are on our third big data stack right now, and I already have a proof of concept running on a fourth stack. I am just starting to look at how we can load balance content traffic across our egress ports, but at the same time improve latency and quality of real time game traffic. In conjunction with this, I am looking at a new feedback system to improve the control over our content delivery and Steam Datagram Relay servers. Streaming telemetry from the network devices makes this much more possible than straight SNMP. On the backburner, I also have a plan for feeding network flow data into a machine learning system to try and identify interesting activity on our network.

Listed skills include Data Center, Networking, Perforce, Visual Effects, and 7 others.