9:00
|
Opening
|
9:15
|
Session: Multipath
Multipath TCP is a TCP extension that enables hosts to send data belonging to a single TCP connection over different paths. It was designed as an incrementally deployable evolution of TCP. For this reason, the Multipath TCP specification assumes that applications use the unmodified socket interface. Given the growing interest in using Multipath TCP for specific applications, there is a demand for an advanced API that enables application developers to control the operation of the Multipath TCP stack. Keeping with the incremental deployment objectives of Multipath TCP, we propose a simple but powerful socket API that uses new socket options to control the operation of the underlying stack. We implement this extension in the reference implementation of Multipath TCP in the Linux kernel and illustrate its usefulness in several use cases.
An enhanced socket API for Multipath TCP.
Full Paper
Benjamin Hesmans (UCL) and Olivier Bonaventure (UCL).
Recent work has applied Multipath TCP proxies to the problem of bonding a customer’s multiple access interfaces to the Internet, in order to augment available bandwidth, especially in areas with marginal fixed connectivity. However, such proxies only apply to TCP traffic, and while UDP-based media streams can be tunneled through bonded TCP connections, this would lose the advantages of loss-tolerant media-oriented transports. We therefore propose an approach to do interface bonding at layer 3, design a scheduling algorithm to shift traffic between fixed and mobile lines, implemented Linux-based bonding
gateways, and tested them within a testbed on Swisscom’s production DSL and LTE networks.
Multipath bonding at Layer 3.
Full Paper
Maciej Bednarek (ETH Zurich), Guillermo Barrenetxea Kobas (Swisscom), Mirja Kühlewind (ETH Zurich), and Brian Trammell (ETH Zurich).
Multipath TCP has been recently introduced in order to allow a better resource consumption and user quality-of-experience. This is achieved by allowing a connection between two hosts through multiple subflows. However, with the rise of middleboxes and inherent Internet ossification, the large-scale deployment of this TCP extension is difficult. In particular, a load balancer at the entry point of a data center may forward subflows to different servers, cancelling so the advantages of Multipath TCP.
In this paper, we introduce MpLb, a Multipath TCP aware load balancer that fixes this particular issue without any modification to the Multipath TCP protocol itself. We demonstrate advantages of MpLb through a proof-of-concept.
Towards a Multipath TCP Aware Load Balancer.
Short Paper
Simon Liénardy (Université de Liège) and Benoit Donnet (Université de Liège).
Small companies and branch offices often have bandwidth demands and redundancy
needs that go beyond the commercially available Internet access products in their
price range. One way to overcome this problem is to bundle existing Internet
access products. In effect, they become multi-homed often without running BGP
or even getting an AS number.
Currently, these users rely on proprietary L4 load balancing routers, proprietary
multi-channel VPN routers, or sometimes LISP, to bundle their “cheaper”
Internet access network links, e.g., via (v)DSL, DOCSIS, HSDPA, or LTE. While
most products claim transport-layer transparency they add complexity via
middleboxes, map each TCP connection to a single interface, and have limited
application support. Thus, in this paper we propose an alternative:
Auto-configuration of multiple IPv6 prefixes on a single L2 link. We
discuss how this enables applications to take advantage of combining
multiple access networks at with minimal system changes.
Multi-Homed on a Single Link: Using Multiple IPv6 Access Networks.
Short Paper
Philipp S. Tiesel (TU-Berlin), Bernd May (TU-Berlin), and Anja Feldmann (TU-Berlin).
|
10:25
|
Break
|
10:45
|
Session: SDN, Routing and Peering
Updating data plane state to adapt to dynamic conditions is a fundamental network control operation. Software-Defined Networking (SDN) offers abstractions for updating network state while preserving consistency properties. However, realizing these abstractions in a purely centralized fashion is inefficient, due to the inherent delays between switches and the SDN controller, we argue for delegating the responsibility of coordinated updates to the switches. To make our case, we propose ez-Segway, a mechanism that enables decentralized network updates while preventing forwarding anomalies and avoiding link congestion. In our architecture, the controller is only responsible for computing the intended network configuration. This information is distributed to the switches, which use partial knowledge and direct message passing to efficiently schedule and implement the update. This separation of concerns has the key benefit of improving update performance as the communication and computation bottlenecks at the controller are removed. Our extensive simulations show update speedups up to a factor of 2.
Towards Decentralized Fast Consistent Updates.
Full Paper
Thanh Dang Nguyen (Université catholique de Louvain), Marco Chiesa (Université catholique de Louvain), and Marco Canini (Université catholique de Louvain).
In this paper we define the notion of composition for software-defined network applications and show the theoretical and practical approaches to composition in software-defined networks and explain the challenges associated with it. We explore the feasibility of OpenFlow as an Application Programming Interface (API) for a composition engine and argue that its design as Southbound controller interface makes it unsuitable for this task.
Composition of SDN applications: Options/challenges for real implementations.
Full Paper
Arne Schwabe (University of Paderborn), Pedro A. Aranda Gutiérrez (Telefónica I+D), and Holger Karl (University of Paderborn).
The growing relevance of Internet eXchange Points (IXPs), where an increasing number of networks exchange routing information, poses fundamental questions regarding the privacy guarantees of confidential business information.
To facilitate the exchange of routes among their members, IXPs provide Route Server (RS) services to dispatch the routes according to each member’s export policies. Nowadays, to make use of RSes, these policies must be disclosed to the IXP.
This state of affairs raises privacy concerns among network administrators and even deters some networks from subscribing to RS services. We design SIXPACK (which stands for “Securing Internet eXchange Points Against Curious onlooKers”), a RS service that leverages Secure Multi-Party Computation (SMPC) techniques to keep export policies confidential, while maintaining the same functionalities as today’s RSes. We assess the effectiveness and scalability of our system by evaluating our prototype implementation and using traces of data from one of the largest IXPs in the world.
Towards Securing Internet eXchange Points Against Curious onlooKers.
Short Paper
Marco Chiesa (Université catholique de Louvain), Daniel Demmler (Technische Universität Darmstadt), Marco Canini (Université catholique de Louvain), Michael Schapira (Hebrew University of Jerusalem), and Thomas Schneider (Technische Universität Darmstadt).
We present a method to compute the customer cones of peering networks using PCH data. Our method computes location dependent customer cones (LDCCs) for networks that are present at more than one IXP instead of computing a single customer cone for each network. We use our method to compute 5753 LDCCs for 3290 IXP participants. Our preliminary analysis of the LDCCs reveals that IXP participants often have different customer cones at different locations.
Computing Customer Cones of Peering Networks.
Short Paper
Jinu Susan Varghese (Iowa State University) and Lu Ruan (Iowa State University).
|
11:55
|
Lunch & Posters
|
13:00
|
Session: Transport Quality and “Happy Eyeballs”
The IETF has developed protocols that promote a healthy IPv4 and IPv6 co-existence. The Happy Eyeballs (HE) algorithm, for instance, prevents bad user experience in situations where IPv6 connectivity is broken. Using an active test (happy) that measures TCP connection establishment times, we evaluate the effects of the HE algorithm. The happy test measures against ALEXA top 10K websites from 80 SamKnows probes connected to dual-stacked networks representing 58 different ASes. Using a 3-years long (2013 - 2016) dataset, we show that TCP connect times to popular websites over IPv6 have considerably improved over time. As of May 2016, 18% of these websites are faster over IPv6 with 91% of the rest at most 1 ms slower. The historical trend shows that only around 1% of the TCP connect times over IPv6 were ever above the HE timer value (300 ms), which leaves around 2% chance for IPv4 to win a HE race towards these websites. As such, 99% of these websites prefer IPv6 connections more than 98% of the time. We show that although absolute TCP connect times (in ms) are not that far apart in both address families, HE with a 300 ms timer value tends to prefer slower IPv6 connections in around 90% of the cases. We show that lowering the HE timer value to 150 ms gives us a margin benefit of 10% while retaining same preference levels over IPv6.
Measuring the Effects of Happy Eyeballs.
Full Paper
Vaibhav Bajpai (Jacobs University Bremen) and Jürgen Schönwälder (Jacobs University Bremen).
Concerns have been raised in the past several years that introducing new transport protocols on the Internet has become increasingly difficult, not least because there is no agreed-upon way for a source end host to find out if a transport protocol is supported all the way to a destination peer. A solution to a similar problem—finding out support for IPv6—has been proposed and is currently being deployed: the Happy Eyeballs (HE) mechanism. HE has also been proposed as an efficient way for an application to select an appropriate transport protocol. Still, there are few, if any, performance evaluations of transport HE. This paper demonstrates that transport HE could indeed be a feasible solution to the transport support problem. The paper evaluates HE between TCP and SCTP using TLS encrypted and unencrypted traffic, and shows that although there is indeed a cost in terms of CPU load to introduce HE, the cost is relatively small, especially in comparison with the cost of using TLS encryption. Moreover, our results suggest that HE has a marginal impact on memory usage. Finally, by introducing caching of previous connection attempts, the additional cost of transport HE could be significantly reduced.
On the Cost of Using Happy Eyeballs for Transport Protocol Selection.
Full Paper
Giorgos Papastergiou (Simula Research Laboratory), Karl-Johan Grinnemo (Karlstad University), Anna Brunstrom (Karlstad University), David Ros (Simula Research Laboratory), Michael Tüxen (Fachhochschule Münster), Naeem Khademi (University of Oslo), and Per Hurtig (Karlstad University).
When multiple TCP connections are used between the same host pair, they often share a common bottleneck – especially when they are encapsulated together, e.g. in VPN scenarios. Then, all connections after the first should not have to guess the right initial value for the congestion window, but rather get the appropriate value from other connections. This allows short flows to complete much faster – but it can also lead to large bursts that cause problems on their own. Prior work used timer-based pacing methods to alleviate this problem; we introduce a new algorithm that “paces” packets by instead correctly maintaining the ACK clock, and show its positive impact in combination with a previously presented congestion coupling algorithm.
Start Me Up: Determining and Sharing TCP’s Initial Congestion Window.
Short Paper
Safiqul Islam (University of Oslo) and Michael Welzl (University of Oslo).
|
14:05
|
Session: Measurement
Ever growing demand for network bandwidth makes computer networks an area of constant development and fast adjustments. The steady change makes good performance assessments equally necessary and challenging. This development motivated us to revisit the established benchmarking methodology. We provide an overview of the state-of-the-art in router benchmarking, the currently available benchmarking tools, and challenges for benchmarks. A discussion of benchmarking results for three different devices (routers based on Linux and FreeBSD, and a MikroTik router) reveal different properties currently not covered by standardized benchmarks. We conclude by adding tests to the common benchmarking methodology reflecting these properties to make the results more valuable.
The prototype software implementation of our own benchmarking tool and its measurement reports are publicly available.
Revisiting Benchmarking Methodology for Interconnect Devices.
Full Paper
Daniel Raumer (Technical University of Munich), Sebastian Gallenmüller (Technical University of Munich), Florian Wohlfart (Technical University of Munich), Paul Emmerich (Technical University of Munich), Patrick Werneck (Technical University of Munich), and Georg Carle (Technical University of Munich).
In today’s Internet we see an increasing deployment of middleboxes. While middleboxes provide in-network functionality that is necessary to keep networks manageable and economically viable, any packet mangling – whether essential for the needed functionality or accidental as an unwanted side effect – makes it more and more difficult to deploy new protocols or extensions of existing protocols. For the evolution of the protocol stack, it is important to know which network impairments exist and potentially need to be worked around. While classical network measurement tools are often focused on absolute performance values, we present a new measurement tool, called PATHspider that performs A/B testing between two different protocols or different protocol extension to perform controlled experiments of protocol-dependent connectivity problems as well as differential treatment. PATHspider is a framework for performing and analyzing these measurements, while the actual A/B test can be easily customized. This paper describes the basic design approach and architecture of PATHspider and gives guidance how to use and customize it.
PATHspider: A tool for active measurement of path transparency.
Short Paper
Iain R. Learmonth (University of Aberdeen), Brian Trammell (ETH Zurich), Mirja Kuhlewind (ETH Zurich), and Gorry Fairhurst (University of Aberdeen).
IPv6 activity is commonly reported as a fraction of network traffic per day. Within this traffic, however, are daily and weekly characteristics, driven by non-uniform IPv6 deployment across ISPs and regions. This paper discusses some of the more apparent patterns we observe today.
Diurnal and Weekly Cycles in IPv6 Traffic.
Short Paper
Stephen D. Strowes (Yahoo).
The IP header should be the ideal part of a packet that an end system could use to ask the network for special treatment.
Recently, there has been renewed interest in using bits of this header – e.g. the ECN and the DSCP fields. But can we really use these bits? Or should we try to use other bits? We contribute to the body of work that tries to answer these questions by reporting on IPv4 measurements regarding the DSCP field and the Evil bit. Our findings also confirm recent results on IP Options and ECN.
How to say that you’re special: Can we use bits in the IPv4 header?
Short Paper
Runa Barik (University of Oslo, Norway), Michael Welzl (University of Oslo, Norway), and Ahmed Elmokashfi (Simula Research Laboratory, Norway).
The Internet is full of Middleboxes that change packets and flows. In
fact, there is probably no IP o TCP header that is not affected by at
least one middlebox. Obviously, middleboxes impede path transparency,
i.e., the idea that an exchange of messages results in more or less
the same packets, no matter what path the packets take.
But no one seems to have a truly global view of what middleboxes do
to packets on what Internet paths, which would however be essential
knowledge for new transport protocols to be successfully deployed.
We address these concerns in the MAMI project by building an
observatory of path transparency measurements. The project hosts an
extensive set of path transparency measure-ments—we believe it to
be the first dataset to deal specifically with middlebox
involvement.
In this paper, we describe that Observatory and a number of questions
that we want to address with the data in that Observatory.
Eventually, the project will provide public access to that
Observatory so that researchers and the interested public can ask
their own questions about path transparency issues and middlebox
involvement.
Towards an Observatory for Network Transparency Research.
Short Paper
Stephan Neuhaus (Zürcher Hochschule für Angewandte Wissenschaften), Roman Müntener (Zürcher Hochschule für Angewandte Wissenschaften), Korian Edeline (Université de Liège), Benoit Donnet (Université de Liège), and Elio Gubser (ETH Zurich).
|
14:55
|
Break & Posters
|
15:45
|
Session: Internet Media
Google congestion control (GCC) has been proposed for the case of delay sensitive traffic (i.e. video-conference) in the WebRTC framework. In this paper we analyze the effect of wireless channel outages on the GCC. We have observed that, when a channel outage ends, there are packets that arrive at the receiver as a burst. This behavior impairs the delay-based controller employed by GCC, resulting in throughput degradation. We propose a solution to make GCC robust with respect to channel outages. In particular, by grouping packets that arrive in a burst, the delay-based controller avoids to misinterpret a burst as network congestion. In order to prove the effectiveness of the proposed solution we have carried out a trace-driven experimental evaluation in a loaded Wi-Fi scenario.
Making Google Congestion Control robust over Wi-Fi networks using packet grouping.
Full Paper
Gaetano Carlucci (Politecnico di Bari, Italy), Luca De Cicco (Politecnico di Bari, Italy), Stefan Holmer (Google Inc, Sweden), and Saverio Mascolo (Politecnico di Bari, Italy).
Real-time applications require a set of transport services not currently
provided by widely-deployed transport protocols. Ossification prevents the
deployment of novel protocols, restricting solutions to protocols using
either TCP or UDP as a substrate. We describe the transport services
required by real-time applications. We show that, in the short-term (i.e.,
while UDP is blocked at current levels), TCP offers a feasible substrate
for providing these services. Over the longer term, protocols using UDP
may reduce the number of networks blocking UDP, enabling a shift towards
its use as a demultiplexing layer for novel transport protocols.
Implementing Real-Time Transport Services over an Ossified Network.
Full Paper
Stephen McQuistin (University of Glasgow), Colin Perkins (University of Glasgow), and Marwan Fayed (University of Stirling).
|
16:45
|
Closing Keynote
|
17:15
|
Wrap-Up, Discussion and Closing
|
|
|
19:00
|
IETF Warm-Up! BBQ
Hosted by the Freie Universität Berlin, free but separate
registration required.
|