Computer Communication Review

Cloud data centers host diverse applications, mixing workloads that require small predictable latency with others requiring large sustained throughput. In this environment, today's state-of-the-art TCP protocol falls short. We present measurements of a 6000 server production cluster and reveal impairments that lead to high application latencies, rooted in TCP's demands on the limited buffer space available in data center switches. For example, bandwidth hungry "background" flows build up queues at the switches, and thus impact the performance of latency sensitive "foreground" traffic.

To address these problems, we propose DCTCP, a TCP-like protocol for data center networks. DCTCP leverages Explicit Congestion Notification (ECN) in the network to provide multi-bit feedback to the end hosts. We evaluate DCTCP at 1 and 10Gbps speeds using commodity, shallow buffered switches. We find DCTCP delivers the same or better throughput than TCP, while using 90% less buffer space. Unlike TCP, DCTCP also provides high burst tolerance and low latency for short flows. In handling workloads derived from operational measurements, we found DCTCP enables the applications to handle 10X the current background traffic, without impacting foreground traffic. Further, a 10X increase in foreground traffic does not cause any timeouts, thus largely eliminating incast problems.

The emergence of Information-Centric Networking~(ICN) provides considerable opportunities for context-aware data distribution in the network's forwarding plane. While packet forwarding in classical IP-based networks is basically predetermined by routing, ICN foresees an adaptive forwarding plane considering the requirements of network applications. As research in this area is still at an early stage, most of the work so far focused on providing the basic functionality, rather than on considering the available context information to improve Quality of Service (QoS). This article investigates to which extent existing forwarding strategies take account of the available context information and can therefore increase service quality. The article examines a typical scenario encompassing different user applications (Voice over IP, video streaming, and classical data transfer) with varying demands (context), and evaluates how well the applications' requirements are met by the existing strategies.

Named Data Networking (NDN) is one of five projects funded by the U.S. National Science Foundation under its Future Internet Architecture Program. NDN has its roots in an earlier project, Content-Centric Networking (CCN), which Van Jacobson first publicly presented in 2006. The NDN project investigates Jacobson's proposed evolution from today's host-centric network architecture (IP) to a data-centric network architecture (NDN). This conceptually simple shift has far-reaching implications for how we design, develop, deploy, and use networks and applications. We describe the motivation and vision of this new architecture, and its basic components and operations. We also provide a snapshot of its current design, development status, and research challenges. More information about the project, including prototype implementations, publications, and annual reports, is available on named-data.net.

In Named Data Networking (NDN) architecture, packets carry data names rather than source or destination addresses. This change of paradigm leads to a new data plane: data consumers send out Interest packets, routers forward them and maintain the state of pending Interests, which is used to guide Data packets back to the consumers. NDN routers' forwarding process is able to detect network problems by observing the two-way traffic of Interest and Data packets, and explore multiple alternative paths without loops. This is in sharp contrast to today's IP forwarding process which follows a single path chosen by the routing process, with no adaptability of its own. In this paper we outline the design of NDN's adaptive forwarding, articulate its potential benefits, and identify open research issues.

This writeup presents a critique of the field of "Wireless Sensor Networks (WSNs)". Literature in this domain falls into two main, distinct categories: (1) algorithms or protocols, and (2) applicationcentric system design. A striking observation is that references across these two categories are minimal, and superficial at best. We argue that this is not accidental, and is the result of three main flaws in the former category of work. Going forward, an applicationdriven, bottom-up approach is required for meaningful articulation and subsequent solution of any networking issues in WSNs.

The Differentiated Service (Diff-Serv) architecture [1] advocates a model based on different “granularity” at network edges and within the network. In particular, core routers are only required to act on a few aggregates that are meant to offer a pre-defined set of service levels. The use of aggregation raises a number of questions for end-to-end services, in particular when crossing domain boundaries where policing actions may be applied. This paper focuses on the impact of such policing actions in the context of individual and the bulk services built on top of the Expedited Forwarding (EF) [7] per-hop-behavior (PHB). The findings of this investigation confirm and quantify the expected need for reshaping at network boundaries, and identify a number of somewhat unexpected behaviors. Recommendations are also made for when reshaping is not available.

As data centers become more and more central in Internet communications, both research and operations communities have begun to explore how to better design and manage them. In this paper, we present a preliminary empirical study of end-to-end traffic patterns in data center networks that can inform and help evaluate research and operational approaches. We analyze SNMP logs collected at 19 data centers to examine temporal and spatial variations in link loads and losses. We find that while links in the core are heavily utilized the ones closer to the edge observe a greater degree of loss. We then study packet traces collected at a small number of switches in one data center and find evidence of ON-OFF traffic behavior. Finally, we develop a framework that derives ON-OFF traffic parameters for data center traffic sources that best explain the SNMP data collected for the data center. We show that the framework can be used to evaluate data center traffic engineering approaches. We are also applying the framework to design network-level traffic generators for data centers.

This paper "peeks under the covers" at the subsystems that provide the basic functionality of a leading content delivery network. Based on our experiences in building one of the largest distributed systems in the world, we illustrate how sophisticated algorithmic research has been adapted to balance the load between and within server clusters, manage the caches on servers, select paths through an overlay routing network, and elect leaders in various contexts. In each instance, we first explain the theory underlying the algorithms, then introduce practical considerations not captured by the theoretical models, and finally describe what is implemented in practice. Through these examples, we highlight the role of algorithmic research in the design of complex networked systems. The paper also illustrates the close synergy that exists between research and industry where research ideas cross over into products and product requirements drive future research.

We describe and analyse in details the various factors that influence the convergence time of intradomain link state routing protocols. This convergence time reflects the time required by a network to react to the failure of a link or a router. To characterise the convergence process, we first use detailed measurements to determine the time required to perform the various operations of a link state protocol on currently deployed routers. We then build a simulation model based on those measurements and use it to study the convergence time in large networks. Our measurements and simulations indicate that sub-second link-state IGP convergence can be easily met on an ISP network without any compromise on stability.

Network operators must have control over the flow of traffic into, out of, and across their networks. However, the Border Gateway Protocol (BGP) does not facilitate common traffic engineering tasks, such as balancing load across multiple links to a neighboring AS or directing traffic to a different neighbor. Solving these problems is difficult because the number of possible changes to routing policies is too large to exhaustively test all possibilities, some changes in routing policy can have an unpredictable effect on the flow of traffic, and the BGP decision process implemented by router vendors limits an operator's control over path selection.We propose fundamental objectives for interdomain traffic engineering and specific guidelines for achieving these objectives within the context of BGP . Using routing and traffic data from the AT&T backbone we show how certain BGP policy changes can move traffic in a predictable fashion, despite limited knowledge about the routing policies in neighboring AS's. Then, we show how operators can gain greater flexibility by relaxing some steps in the BGP decision process and ensuring that neighboring AS's send consistent advertisements at each peering location. Finally, we show that an operator can manipulate traffic efficiently by changing the routes for a small number of prefixes (or groups of related prefixes) that consistently receive a large amount of traffic.