Modern data centers host hundreds of applications with hundreds of thousands of flows and varied traffic patterns. Efficient management of the data center flows is critical in order to ensure high utilization of the network and also achieving desired application performance. Numerous proposals aim to improve data center and enterprise networks through better management of network traffic. Examples include load balancing using Equal Cost Multipathing (ECMP), traffic engineering, precise per-flow rate limiting, scalable flow prioritization and classification, and support for microsecond-scale TDMA. These proposals rely on (1) support from the switching hardware for in-network flow management and/or (2) require the end host data plane being sufficiently performant and high precision, in order to work in coordination with network switches and centralized controllers. While these flow management proposals meet the high resource utilization and performance requirement, often they rely on hardware modifications making them hard to deploy or cost- prohibitive, or, inextensible interfaces that make it hard for them to inter- operate with the existing software stack in the end hosts. To address these challenges, we evaluate and propose readily deployable, cost-effective mechanisms for flow management both within the data center network fabric, leveraging commodity switch hardware and at the end host stack by leveraging the high performance packet processing libraries with better interfacing with the applications and the network. We propose Weighted Cost Multipathing (WCMP) as an in-network flow management framework that allows network operators to manage flows and load balance traffic with existing commodity switches. WCMP requires no hardware modification and allows fair bandwidth sharing between flows in presence of failures and asymmetry deliver high utilization and application performance. For end host based flow management, we evaluate the existing state-of-art mechanisms for flow management and leverage the high performance packet processing frameworks such as Dataplane development kit (DPDK) to propose a software defined dataplane (SDD) that allows the end host stack to interface with a remote controller and provide fine-grain control over packet transmissions and flow management

Solving "Big Data" problems requires bridging massive quantities of compute, memory, and storage, which requires significant amounts of bisection bandwidth. Topologies like Fat-tree, VL2, and HyperX achieve a scale-out design by leveraging multiple paths from source to destination. However, traditional routing protocols are not able to effectively utilize these links while also harnessing spare network capacity to better statistically multiplex network resources. In this work we present Dahu, a switch mechanism for efficiently load balancing traffic in multipath networks. Dahu avoids congestion hot-spots by dynamically spreading traffic uniformly across links, and forwarding traffic over non-minimal paths where possible. By performing load balancing primarily using local information, Dahu can act more quickly than centralized approaches, and responds to failure gracefully. Our evaluation shows that Dahu delivers up to 50% higher throughput relative to ECMP in an 8,192 server Fat-tree network and up to 500% improvement in throughput in large scale HyperX networks with over 130,000 servers.

Pre-2018 CSE ID: CS2013-0992

In datacenter applications, predictability in service time and controlled latency, especially tail latency, is essential for building performant applications. This is especially true for applications or services build by accessing data across thousands of servers to generate a user response. Current practice has been to run such services at low utilization to rein in latency outliers, which decreases efficiency and limits the number of service invocations developers can issue while still meeting tight latency budgets. In this paper, we analyze the three datacenter applications, Memcached, OpenFlow, and web search, to measure the effect on tail latency of 1) kernel socket handling, NIC interaction, and the network stack, 2) application locks contested in the kernel, and 3) application-layer queuing due to requests being stalled behind straggler threads. We propose a novel approach of reducing the above sources of latency by relying on support from the NIC hardware, and we find that the resulting improvements indeed dramatically reduce end-to-end application latency.

Pre-2018 CSE ID: CS2012-0977