Fabián E. Bustamante

Taming the torrent: a practical approach to reducing cross-isp traffic in peer-to-peer systems

Peer-to-peer (P2P) systems, which provide a variety of popular services, such as file sharing, video streaming and voice-over-IP, contribute a significant portion of todays Internet traffic. By building overlay networks that are oblivious to the underlying Internet topology and routing, these systems have become one of the greatest traffic-engineering challenges for Internet Service Providers (ISPs) and the source of costly data traffic flows. In an attempt to reduce these operational costs, ISPs have tried to shape, block or otherwise limit P2P traffic, much to the chagrin of their subscribers, who consistently finds ways to eschew these controls or simply switch providers. In this paper, we present the design, deployment and evaluation of an approach to reducing this costly cross-ISP traffic without sacrificing system performance. Our approach recycles network views gathered at low cost from content distribution networks to drive biased neighbor selection without any path monitoring or probing. Using results collected from a deployment in BitTorrent with over 120,000 users in nearly 3,000 networks, we show that our lightweight approach significantly reduces cross-ISP traffic and, over 33% of the time, it selects peers along paths that are within a single autonomous system (AS). Further, we find that our system locates peers along paths that have two orders of magnitude lower latency and 30% lower loss rates than those picked at random, and that these high-quality paths can lead to significant improvements in transfer rates. In challenged settings where peers are overloaded in terms of available bandwidth, our approach provides 31% average download-rate improvement; in environments with large available bandwidth, it increases download rates by 207% on average (and improves median rates by 883%

An integrated mobility and traffic model for vehicular wireless networks

Ad-hoc wireless communication among highly dynamic, mobile nodes in a urban network is a critical capability for a wide range of important applications including automated vehicles, real-time traffic monitoring and vehicular safety applications. When evaluating application performance in simulation, a realistic mobility model for vehicular ad-hoc networks (VANETs) is critical for accurate results. This paper analyzes ad-hoc wireless network performance in a vehicular network in which nodes move according to a simplified vehicular traffic model on roads defined by real map data. We show that when nodes move according to our street mobility model, STRAW, network performance is significantly different from that of the commonly used random waypoint model. We also demonstrate that protocol performance varies with the type of urban environment. Finally, we use these results to argue for the development of integrated vehicular and network traffic simulators to evaluate vehicular ad-hoc network applications, particularly when the information passed through the network affects node mobility.

Drafting behind Akamai: inferring network conditions based on CDN redirections

To enhance Web browsing experiences, content distribution networks (CDNs) move Web content ¿closer¿ to clients by caching copies of Web objects on thousands of servers worldwide. Additionally, to minimize client download times, such systems perform extensive network and server measurements and use them to redirect clients to different servers over short time scales. In this paper, we explore techniques for inferring and exploiting network measurements performed by the largest CDN, Akamai; our objective is to locate and utilize quality Internet paths without performing extensive path probing or monitoring. Our contributions are threefold. First, we conduct a broad measurement study of Akamais CDN. We probe Akamais network from 140 PlanetLab (PL) vantage points for two months. We find that Akamai redirection times, while slightly higher than advertised, are sufficiently low to be useful for network control. Second, we empirically show that Akamai redirections overwhelmingly correlate with network latencies on the paths between clients and the Akamai servers. Finally, we illustrate how large-scale overlay networks can exploit Akamai redirections to identify the best detouring nodes for one-hop source routing. Our research shows that in more than 50% of investigated scenarios, it is better to route through the nodes ¿recommended¿ by Akamai than to use the direct paths. Because this is not the case for the rest of the scenarios, we develop low-overhead pruning algorithms that avoid Akamai-driven paths when they are not beneficial. Because these Akamai nodes are part of a closed system, we provide a method for mapping Akamai-recommended paths to those in a generic overlay and demonstrate that these one-hop paths indeed outperform direct ones.

Where the sidewalk ends: extending the internet as graph using traceroutes from P2P users

An accurate Internet topology graph is important in many areas of networking, from understanding ISP business relationships to diagnosing network anomalies. Most Internet mapping efforts have derived the network structure, at the level of interconnected autonomous systems (ASes), from a rather limited set of vantage points. In this paper, we argue that a promising approach to revealing the hidden areas of the Internet topology is through active measurement from an observation platform that scales with the growing Internet. By leveraging measurements performed by an extension to a popular P2P system, we show that this approach indeed exposes significant new topological information. Our study is based on traceroute measurements from more than 992,000 IPs in over 3,700 ASes distributed across the Internet hierarchy, many in regions of the Internet not covered by publicly available path information. To address this issue we develop heuristics that identify 23,914 new AS links not visible in the publicly-available BGP data-12.86 percent more customer-provider links and 40.99 percent more peering links, than previously reported. We validate our heuristics using data from a tier-1 ISP, and show that they successfully filter out all false links introduced by public IP-to-AS mapping. We analyze properties of the Internet graph that includes these new links and characterize why they are missing. Finally, we have made the identified set of links and their inferred relationships publicly available.

Content delivery and the natural evolution of DNS: remote dns trends, performance issues and alternative solutions

Content Delivery Networks (CDNs) rely on the Domain Name System (DNS) for replica server selection. DNS-based server selection builds on the assumption that, in the absence of information about the clients actual network location, the location of a clients DNS resolver provides a good approximation. The recent growth of remote DNS services breaks this assumption and can negatively impact clients web performance. In this paper, we assess the end-to-end impact of using remote DNS services on CDN performance and present the first evaluation of an industry-proposed solution to the problem. We find that remote DNSusage can indeed significantly impact clients web performance and that the proposed solution, if available, can effectively address the problem for most clients. Considering the performance cost of remote DNS usage and the limited adoption base of the industry-proposed solution, we present and evaluate an alternative approach, Direct Resolution, to readily obtain comparable performance improvements without requiring CDN or DNS participation.

Characterizing and Predicting TCP Throughput on the Wide Area Network

DualPats exploits the strong correlation between TCP throughput and flow size, and the statistical stability of Internet path characteristics to accurately predict the TCP throughput of large transfers using active probing. We propose additional mechanisms to explain the correlation, and then analyze why traditional TCP benchmarking fails to predict the throughput of large transfers well. We characterize stability and develop a dynamic sampling rate adjustment algorithm so that we probe a path based on its stability. Our analysis, design, and evaluation is based on a large-scale measurement study

FatNemo: Building a Resilient Multi-source Multicast Fat-Tree

This paper proposes the idea of emulating fat-trees in overlays for multi-source multicast applications. Fat-trees are like real trees in that their branches become thicker the closer one gets to the root, thus overcoming the “root bottleneck” of regular trees. We introduce FatNemo, a novel overlay multi-source multicast protocol based on this idea. FatNemo organizes its members into a tree of clusters with cluster sizes increasing closer to the root. It uses bandwidth capacity to decide the highest layer in which a peer can participate, and relies on co-leaders to share the forwarding responsibility and to increase the tree’s resilience to path and node failures.

Crowdsourcing ISP characterization to the network edge

Evaluating and characterizing Internet Service Providers (ISPs) is critical to subscribers shopping for alternative ISPs, companies providing reliable Internet services, and governments surveying the coverage of broadband services to its citizens. Ideally, ISP characterization should be done at scale, continuously, and from end users. While there has been significant progress toward this end, current approaches exhibit apparently unavoidable tradeoffs between coverage, continuous monitoring and capturing user-perceived performance. In this paper, we argue that network-intensive applications running on end systems avoid these tradeoffs, thereby offering an ideal platform for ISP characterization. Based on data collected from 500,000 peer-to-peer BitTorrent users across 3,150 networks, together with the reported results from the U.K. Ofcom/SamKnows studies, we show the feasibility of this approach to characterize the service that subscribers can expect from a particular ISP. We discuss remaining research challenges and design requirements for a solution that enables efficient and accurate ISP characterization at an Internet scale.

Resilient peer-to-peer multicast without the cost

We introduce Nemo, a novel peer-to-peer multicast protocol that achieves high delivery ratio without sacrificing end-to-end latency or incurring additional costs. Based on two simple techniques: (1) co-leaders to minimize dependencies and, (2) triggered negative acknowledgments (NACKs) to detect lost packets, Nemos design emphasizes conceptual simplicity and minimum dependencies, thus achieving performance characteristics capable of withstanding the natural instability of its target environment. We present an extensive comparative evaluation of our protocol through simulation and wide-area experimentation. We contrast the scalability and performance of Nemo with that of three alternative protocols: Narada, Nice and Nice-PRM. Our results show that Nemo can achieve delivery ratios similar to those of comparable protocols under high failure rates, but at a fraction of their cost in terms of duplicate packets (reductions > 90%) and control-related traffic.

Native data representation: An efficient wire format for high-performance distributed computing

New trends in high-performance software development such as tool- and component-based approaches have increased the need for flexible and high-performance communication systems. When trying to reap the well-known benefits of these approaches, the question of what communication infrastructure should be used to link the various components arises. In this context, flexibility and high-performance seem to be incompatible goals. Traditional HPC-style communication libraries, such as MPI, offer good performance, but are not intended for loosely-coupled systems. Object- and metadata-based approaches like XML offer the needed plug-and-play flexibility, but with significantly lower performance. We observe that the flexibility and baseline performance of data exchange systems are strongly determined by their wire formats, or by how they represent data for transmission in heterogeneous environments. After examining the performance implications of using a number of different wire formats, we propose an alternative approach for flexible high-performance data exchange, Native Data Representation, and evaluate its current implementation in the portable binary I/O library.