Conference Program

Abstract:

Low-Power Wide Area Networks (LP-WANs) are an attractive emerging platform to connect the Internet-of-things. LP-WANs enable low-cost devices with a 10-year battery to communicate at few kbps to a base station, kilometers away. But deploying LP-WANs in large urban environments is challenging, given the sheer density of nodes that causes interference, coupled with attenuation from buildings that limits signal range. Yet, state-of-the-art techniques to address these limitations demand inordinate hardware complexity at the base stations or clients, increasing their size and cost.

This paper presents Choir, a system that overcomes challenges pertaining to density and range of urban LP-WANs despite the limited capabilities of base station and client hardware. First, Choir proposes a novel technique that aims to disentangle and decode large numbers of interfering transmissions at a simple, single-antenna LP-WAN base station. It does so, perhaps counter-intuitively, by taking the hardware imperfections of low-cost LP-WAN clients to its advantage. Second, Choir exploits the correlation of sensed data collected by LP-WAN nodes to collaboratively reach a far-away base station, even if individual clients are beyond its range. We implement and evaluate Choir on USRP N210 base stations serving a 10 square kilometer area surrounding Carnegie Mellon University campus. Our results reveal that Choir improves network throughput of commodity LP-WAN clients by 6.84× and expands communication range by 2.65×.

Abstract:

This paper presents the design and implementation of Wi-Fi Goes to Town, the first Wi-Fi based roadside hotspot network designed to operate at vehicular speeds with meter-sized picocells. Wi-Fi Goes to Town APs make delivery decisions to the vehicular clients they serve at millisecond-level granularities, exploiting path diversity in roadside networks. In order to accomplish this, we introduce new buffer management algorithms that allow participating APs to manage each others’ queues, rapidly quenching each others’ transmissions and flushing each others’ queues. We furthermore integrate our fine-grained AP selection and queue management into 802.11’s frame aggregation and block acknowledgment functions, making the system effective at modern 802.11 bit rates that need frame aggregation to maintain high spectral efficiency. We have implemented our system in an eight-AP network alongside a nearby road, and evaluate its performance with mobile clients moving at up to 35 mph. Depending on the clients’ speed, Wi-Fi Goes to Town achieves a 2.4–4.7X TCP throughput improvement over a baseline fast handover protocol that captures the state of the art in Wi-Fi roaming, including the recent IEEE 802.11k and 802.11r standards.

Abstract:

Battery-free sensors, such as RFIDs, are annually attached to billions of items including pharmaceutical drugs, clothes, and manufacturing parts. The fundamental challenge with battery-free sensors is that they are only reliable at short distances of tens of centimeters to few meters. As a result, today’s systems for communicating with and localizing battery-free sensors are crippled by the limited range.

To overcome this challenge, this paper presents RFly, a system that leverages drones as relays for battery-free networks. RFly delivers two key innovations. It introduces the first full-duplex relay for battery-free networks. The relay can seamlessly integrate with a deployed RFID infrastructure, and it preserves phase and timing characteristics of the forwarded packets. RFly also develops the first RF-localization algorithm that can operate through a mobile relay.

We built a hardware prototype of RFly’s relay into a custom PCB circuit and mounted it on a Parrot Bebop drone. Our experimental evaluation demonstrates that RFly enables communication with commercial RFIDs at over 50m. Moreover, its through-relay localization algorithm has a median accuracy of 19 centimeters. These results demonstrate that RFly provides powerful primitives for communication and localization in battery-free networks.

Abstract:

Cellular traffic continues to grow rapidly making the scalability of the cellular infrastructure a critical issue. However, there is mounting evidence that the current Evolved Packet Core (EPC) is ill-suited to meet these scaling demands: EPC solutions based on specialized appliances are expensive to scale and recent software EPCs perform poorly, particularly with increasing numbers of devices or signaling traffic.

In this paper, we design and evaluate a new system architecture for a software EPC that achieves high and scalable performance. We postulate that the poor scaling of existing EPC systems stems from the manner in which the system is decomposed which leads to device state being duplicated across multiple components which in turn results in frequent interactions between the different components. We propose an alternate approach in which state for a single device is consolidated in one location and EPC functions are (re)organized for efficient access to this consolidated state. In effect, our design “slices” the EPC by user.

We prototype and evaluate PEPC, a software EPC that implements the key components of our design. We show that PEPC achieves 3-7x higher throughput than comparable software EPCs that have been implemented in industry and over 10x higher throughput than a popular open-source implementation (OpenAirInterface). Compared to the industrial EPC implementations, PEPC sustains high data throughput for 10-100x more users devices per core, and a 10x higher ratio of signaling-to-data traffic. In addition to high performance, PEPC’s by-user organization enables efficient state migration and customization of processing pipelines. We implement user migration in PEPC and show that state can be migrated with little disruption, e.g., migration adds only up to 4us of latency to median per packet latencies.

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%.

Abstract:

We present the design, implementation, validation, and deployment of the Price $heriff, a highly distributed system for detecting various types of online price discrimination in e-commerce. The Price $heriff uses a peer-to-peer architecture, sandboxing, and secure multiparty computation to allow users to tunnel price check requests through the browsers of other peers without tainting their local or server-side browsing history and state. Having operated the Price $heriff for several months with approximately one thousand real users, we identify several instances of cross-border price discrimination based on the country of origin. Even within national borders, we identify several retailers that return different prices for the same product to different users. We examine whether the observed differences are due to personal-data-induced discrimination or A/B testing, and conclude that it is the latter.

Abstract:

The existing slowness of the web on mobile devices frustrates users and hurts the revenue of website providers. Prior studies have attributed high page load times to dependencies within the page load process: network latency in fetching a resource delays its processing, which in turn delays when dependent resources can be discovered and fetched.

To securely address the impact that these dependencies have on page load times, we present Vroom, a rethink of how clients and servers interact to facilitate web page loads. Unlike existing solutions, which require clients to either trust proxy servers or discover all the resources on any page themselves, Vroom’s key characteristics are that clients fetch every resource directly from the domain that hosts it but web servers aid clients in discovering resources. Input from web servers decouples a client’s processing of resources from its fetching of resources, thereby enabling independent use of both the CPU and the network. As a result, Vroom reduces the median page load time by more than 5 seconds across popular News and Sports sites. To enable these benefits, our contributions lie in making web servers capable of accurately aiding clients in resource discovery and judiciously scheduling a client’s receipt of resources.

Abstract:

Traffic shaping, including pacing and rate limiting, is fundamental to the correct and efficient operation of both datacenter and wide area networks. Sample use cases include policy-based bandwidth allocation to flow aggregates, rate-based congestion control algorithms, and packet pacing to avoid bursty transmissions that can overwhelm router buffers. Driven by the need to scale to millions of flows and to apply complex policies, traffic shaping is moving from network switches into the end hosts, typically implemented in software in the kernel networking stack.

In this paper, we show that the performance overhead of end-host traffic shaping is substantial limits overall system scalability as we move to thousands of individual traffic classes per server. Measurements from production servers show that shaping at hosts consumes considerable CPU and memory, unnecessarily drops packets, suffers from head of line blocking and inaccuracy, and does not provide backpressure up the stack. We present Carousel, a framework that scales to tens of thousands of policies and flows per server, built from the synthesis of three key ideas: i) a single queue shaper using time as the basis for releasing packets, ii) fine-grained, just-in-time freeing of resources in higher layers coupled to actual packet departures, and iii) one shaper per CPU core, with lock-free coordination. Our production experience in serving video traffic at a Cloud service provider shows that Carousel shapes traffic accurately while improving overall machine CPU utilization by 8% (an improvement of 20% in the CPU utilization attributed to networking) relative to state-of-art deployments. It also conforms 10 times more accurately to target rates, and consumes two orders of magnitude less memory than existing approaches.

Abstract:

 Large content providers build points of presence around the world, each connected to tens or hundreds of networks. Ideally, this connectivity lets providers better serve users, but providers cannot obtain enough capacity on some preferred peering paths to handle peak traffic demands. These capacity constraints, coupled with volatile traffic and performance and the limitations of the 20 year old BGP protocol, make it difficult to best use this connectivity.    We present Edge Fabric, an SDN-based system we built and deployed to tackle these challenges for Facebook, which serves over two billion users from dozens of points of presence on six continents. We provide the first public details on the connectivity of a provider of this scale, including opportunities and challenges. We describe how Edge Fabric operates in near real-time to avoid congesting links at the edge of Facebook’s network. Our evaluation on production traffic worldwide demonstrates that Edge Fabric efficiently uses interconnections without congesting them and degrading performance. We also present real-time performance measurements of available routes and investigate incorporating them into routing decisions. We relate challenges, solutions, and lessons from four years of operating and evolving Edge Fabric.

Abstract:

We present the design of Espresso, Google’s SDN-based Internet peering edge routing infrastructure. This architecture grew out of a need to exponentially scale the Internet edge cost-effectively and to enable application-aware routing at Internet-peering scale. Espresso utilizes commodity switches and host-based routing/packet processing to implement a novel fine-grained traffic engineering capability. Overall, Espresso provides Google a scalable peering edge that is programmable, reliable, and integrated with global traffic systems. Espresso also greatly accelerated deployment of new networking features at our peering edge. Espresso has been in production for two years and serves over 22% of Google’s total traffic to the Internet.

Abstract:

Peering infrastructures, namely, colocation facilities and Internet exchange points, are located in every major city, have hundreds of network members, and support hundreds of thousands of interconnections around the globe. These infrastructures are well provisioned and managed, but outages have to be expected, e.g., due to power failures, human errors, attacks, and natural disasters. However, little is known about the frequency and impact of outages at these critical infrastructures with high peering concentration. In this paper, we develop a novel and lightweight methodology for detecting peering infrastructure outages. Our methodology relies on the observation that BGP communities, announced with routing updates, are an excellent and yet unexplored source of information allowing us to pinpoint outage locations with high accuracy. We build and operate a system that can locate the epicenter of infrastructure outages at the level of a building and track the reaction of networks in near real-time. Our analysis unveils four times as many outages as compared to those publicly reported over the past five years. Moreover, we show that such outages have significant impact on remote networks and peering infrastructures. Our study provides a unique view of the Internet’s behavior under stress that often goes unreported.

Abstract:

Network operators often face the problem of remote transit outages, which can cause data-plane downtimes. The key issue is that BGP, the Internet routing protocol, converges very slowly upon such outages, as large bursts of messages have to be processed and propagated router-by-router.

In this paper, we present SWIFT, a fast-reroute mechanism which enables routers to restore connectivity in few seconds upon remote transit outages. SWIFT is based on two novel techniques. First, SWIFT deals with slow outage notification by predicting the overall extent of a remote failure out of few control-plane (BGP) messages. The key insight is that significant inference speed can be gained at the price of some accuracy. Second, SWIFT introduces a new data-plane encoding scheme which enables quick and flexible update of the impacted forwarding entries. SWIFT is deployable on existing devices, without modifying BGP.

We implemented SWIFT and demonstrate that it is both fast and accurate. It predicts the extent of a remote outage in few seconds with an accuracy of ~90% and can restore connectivity for 99% of the impacted destinations.

Abstract:

The Internet’s inter-domain routing infrastructure, provided today by BGP, is extremely rigid and does not facilitate the introduction of new inter-domain routing protocols. This rigidity has made it incredibly difficult to widely deploy critical fixes to BGP. It has also depressed ASes’ ability to sell value-added services or replace BGP entirely with a more sophisticated protocol. Even if operators undertook the significant effort needed to fix or replace BGP, it is likely the next protocol will be just as difficult to change or evolve. To help, this paper identifies two features needed in the routing infrastructure (i.e., within any inter-domain routing protocol) to facilitate evolution to new protocols. To understand their utility, it presents D-BGP, a version of BGP that incorporates them.

Abstract:

We propose and evaluate a new metric for understanding the dependence of the AS-level Internet on individual routers. Whereas prior work uses large volumes of reachability probes to infer outages, we design an efficient active probing technique that directly and unambiguously reveals router restarts. We use our technique to survey 149,560 routers across the Internet for 2.5 years. 59,175 of the surveyed routers (40%) experience at least one reboot, and we quantify the resulting impact of each router outage on global IPv4 and IPv6 BGP reachability.

Our technique complements existing data and control plane outage analysis methods by providing a causal link from BGP reachability failures to the responsible router(s) and multi-homing configurations. While we found the Internet core to be largely robust, we identified specific routers that were single points of failure for the prefixes they advertised. In total, 2,385 routers – 4.0% of the routers that restarted over the course of 2.5 years of probing – were single points of failure for 3,396 IPv6 prefixes announced by 1,708 ASes. We inferred 59% of these routers were the customer-edge border router. 2,374 (70%) of the withdrawn prefixes were not covered by a less specific prefix, so 1,726 routers (2.9%) of those that restarted were single points of failure for at least one network. However, a covering route did not imply reachability during a router outage, as no previously-responsive address in a withdrawn more specific prefix responded during a one-week sample. We validate our reboot and single point of failure inference techniques with four networks, finding no false positive or false negative reboots, but find some false negatives in our single point of failure inferences.

Abstract: In the past five years, the graduate networking course at Stanford has assigned over 200 students the task of reproducing results from over 40 networking papers. We began the project as a means of teaching both engineering rigor and critical thinking, qualities that are necessary for careers in networking research and industry. We have observed that reproducing research can simultaneously be a tool for education and a means for students to contribute to the networking community. Through this editorial we describe our project in reproducing network research and show through anecdotal evidence that this project is important for both the classroom and the networking community at large, and we hope to encourage other institutions to host similar class projects.

Abstract:

Measurement has become fundamental to the operation of networks and at-scale services—whether for management, security, diagnostics, optimization, or simply enhancing our collective understanding of the Internet as a complex system. Further, measurements are useful across points of view— from end hosts to enterprise networks and data centers to the wide area Internet. We observe that many measurements are decoupled from the protocols and applications they are designed to illuminate. Worse, current measurement practice often involves the exploitation of side-effects and unintended features of the network; or, in other words, the artful piling of hacks atop one another. This state of affairs is a direct result of the relative paucity of diagnostic and measurement capabilities built into today’s network stack.

Given our modern dependence on ubiquitous measurement, we propose measurability as an explicit low-level goal of current protocol design, and argue that measurements should be available to all network protocols throughout the stack. We seek to generalize the idea of measurement within protocols, e.g., the way in which TCP relies on measurement to drive its end-to-end behavior. Rhetorically, we pose the question: what if the stack had been built with measurability and diagnostic support in mind? We start from a set of principles for explicit measurability, and define primitives that, were they supported by the stack, would not only provide a solid foundation for protocol design going forward, but also reduce the cost and increase the accuracy of measuring the network.