We report on experiences with Swift congestion control in Google datacenters. Swift targets an end-to-end delay by using AIMD control, with pacing under extreme congestion. With accurate RTT measurement and care in reasoning about delay targets, we find this design is a foundation for excellent performance when network distances are well-known. Importantly, its simplicity helps us to meet operational challenges. Delay is easy to decompose into fabric and host components to separate concerns, and effortless to deploy and maintain as a congestion signal while the datacenter evolves. In large-scale testbed experiments, Swift delivers a tail latency of <50μs for short RPCs, with near-zero packet drops, while sustaining ~100Gbps throughput per server. This is a tail of <3x the minimal latency at a load close to 100%. In production use in many different clusters, Swift achieves consistently low tail completion times for short RPCs, while providing high throughput for long RPCs. It has loss rates that are at least 10x lower than a DCTCP protocol, and handles O(10k) incasts that sharply degrade with DCTCP.

This paper presents our design and experience with a microkernel-inspired approach to host networking called Snap. Snap is a userspace networking system that supports Google's rapidly evolving needs with flexible modules that implement a range of network functions, including edge packet switching, virtualization for our cloud platform, traffic shaping policy enforcement, and a high-performance reliable messaging and RDMA-like service. Snap has been running in production for over three years, supporting the extensible communication needs of several large and critical systems.

The increasing density of WiFi access points (APs) in metropolitan areas is enabling an opportunistic model of wireless networking, whereby a "guest" user within range of one or more wireless APs can gain temporary Internet access through these APs. In this paper, we address the problem of TCP throughput prediction for opportunistic networks. Applications of opportunistic networking can benefit from such predictions by adapting to prevailing network conditions. Our approach is different from prior efforts to model wireless network throughput in that only the two communicating endpoints participate in the prediction, and no information about network topology or traffic loads generated by interfering sources is required. Our goal is to understand how accurate throughput predictions can be under the above assumptions. The physical environment considered in our study includes varying degrees of interference, indoor and outdoor networks, and nodes that are stationary or moving at walking or driving speeds. We use throughput predictors based on time series analysis and machine learning techniques, as they are well-suited to predicting phenomena with unknown variables. The prediction accuracy that our methods yield is cause for cautious optimism. We find that 80% to 100% of predictions are within a factor of two of actual throughput. This bound on accuracy means that predictions are useful for certain applications, because this bound (a) can be achieved by measurements lasting for as little as 0.3 seconds, and (b) holds even when nodes are driving at speeds of 15-25 mph.