Showing papers by "Thomas Anderson published in 2017"

Strata: A Cross Media File System

Abstract: Current hardware and application storage trends put immense pressure on the operating system's storage subsystem. On the hardware side, the market for storage devices has diversified to a multi-layer storage topology spanning multiple orders of magnitude in cost and performance. Above the file system, applications increasingly need to process small, random IO on vast data sets with low latency, high throughput, and simple crash consistency. File systems designed for a single storage layer cannot support all of these demands together. We present Strata, a cross-media file system that leverages the strengths of one storage media to compensate for weaknesses of another. In doing so, Strata provides performance, capacity, and a simple, synchronous IO model all at once, while having a simpler design than that of file systems constrained by a single storage device. At its heart, Strata uses a log-structured approach with a novel split of responsibilities among user mode, kernel, and storage layers that separates the concerns of scalable, high-performance persistence from storage layer management. We quantify the performance benefits of Strata using a 3-layer storage hierarchy of emulated NVM, a flash-based SSD, and a high-density HDD. Strata has 20-30% better latency and throughput, across several unmodified applications, compared to file systems purpose-built for each layer, while providing synchronous and unified access to the entire storage hierarchy. Finally, Strata achieves up to 2.8x better throughput than a block-based 2-layer cache provided by Linux's logical volume manager.

Evaluating the power of flexible packet processing for network resource allocation

Abstract: Recent hardware switch architectures make it feasible to perform flexible packet processing inside the network. This allows operators to configure switches to parse and process custom packet headers using flexible match+action tables in order to exercise control over how packets are processed and routed. However, flexible switches have limited state, support limited types of operations, and limit per-packet computation in order to be able to operate at line rate. Our work addresses these limitations by providing a set of general building blocks that mask these limitations using approximation techniques and thereby enabling the implementation of realistic network protocols. In particular, we use these building blocks to tackle the network resource allocation problem within datacenters and realize approximate variants of congestion control and load balancing protocols, such as XCP, RCP, and CONGA, that require explicit support from the network. Our evaluations show that these approximations are accurate and that they do not exceed the hardware resource limits associated with these flexible switches. We demonstrate their feasibility by implementing RCP with the production Cavium CNX880xx switch. This implementation provides significantly faster and lower-variance flow completion times compared with TCP.

Understanding and Mitigating Packet Corruption in Data Center Networks

Abstract: We take a comprehensive look at packet corruption in data center networks, which leads to packet losses and application performance degradation. By studying 350K links across 15 production data centers, we find that the extent of corruption losses is significant and that its characteristics differ markedly from congestion losses. Corruption impacts fewer links than congestion, but imposes a heavier loss rate; and unlike congestion, corruption rate on a link is stable over time and is not correlated with its utilization.Based on these observations, we developed CorrOpt, a system to mitigate corruption. To minimize corruption losses, it intelligently selects which corrupting links can be safely disabled, while ensuring that each top-of-rack switch has a minimum number of paths to reach other switches. CorrOpt also recommends specific actions (e.g., replace cables, clean connectors) to repair disabled links, based on our analysis of common symptoms of different root causes of corruption. Our recommendation engine has been deployed in over seventy data centers of a large cloud provider. Our analysis shows that, compared to current state of the art, CorrOpt can reduce corruption losses by three to six orders of magnitude and improve repair accuracy by 60%.

Delta-net: Real-time Network Verification Using Atoms

Abstract: Real-time network verification promises to automatically detect violations of network-wide reachability invariants on the data plane. To be useful in practice, these violations need to be detected in the order of milliseconds, without raising false alarms. To date, most real-time data plane checkers address this problem by exploiting at least one of the following two observations: (i) only small parts of the network tend to be affected by typical changes to the data plane, and (ii) many different packets tend to share the same forwarding behaviour in the entire network. This paper shows how to effectively exploit a third characteristic of the problem, namely: similarity among forwarding behaviour of packets through parts of the network, rather than its entirety. We propose the first provably amortized quasi-linear algorithm to do so. We implement our algorithm in a new real-time data plane checker, Delta-net. Our experiments with SDN-IP, a globally deployed ONOS software-defined networking application, and several hundred million IP prefix rules generated using topologies and BGP updates from real-world deployed networks, show that Delta-net checks a rule insertion or removal in approximately 40 microseconds on average, a more than 10X improvement over the state-of-the-art. We also show that Delta-net eliminates an inherent bottleneck in the state-of-the-art that restricts its use in answering Datalog-style "what if" queries.

RAIL: A Case for Redundant Arrays of Inexpensive Links in Data Center Networks.

Abstract: While there are many proposals to reduce the cost of data center networks (DCN), little attention has been paid to the role played by the physical links that carry packets. By studying over 300K optical links across many production DCNs, we show that these links are operating quite conservatively relative to the requirements in the IEEE standards. Motivated by this observation, to reduce DCN costs, we propose using transceivers—a key contributor to DCN cost—beyond their currently specified limit. Our experiments with multiple commodity transceivers show that their reach can be “stretched” 1.6 to 4 times their specification. However, with stretching, the performance of 1–5% of the DCN paths can fall below the IEEE standard. We develop RAIL, a system to ensure that in such a network, applications only use paths that meet their performance needs. Our proposal can reduce the network cost by up to 10% for 10Gbps networks and 44% for 40Gbps networks, without affecting the applications’ performance.