In this paper, we study the effect of reducing the use of BGP’s MRAI mechanism in the Internet. MRAI stands for Minimum Route Advertisement Interval and it is a timer that rate-limits BGP route announcements. In particular, we look into the effects of MRAI removal by simulating router topologies representing all the Autonomous Systems (ASes) in the Internet. We consider the two main MRAI implementation flavors, namely per-session and out-delay MRAI. Reducing the fraction of MRAI-enabled ASes is positive in terms of convergence performance, until the adoption reaches 10% of Internet ASes. Below this threshold, the number of messages exchanged when routes are removed sharply increases, and thus the router load. Depending on the time that it takes for routers to process BGP messages, further reductions in the fraction may even increase convergence time for removing a route. We show that keeping MRAI enabled in few highly connected ASes, the tier-1 ASes, provides some protection against the increase in the computing demand of the routers.

LEDBAT++ is a congestion-control algorithm that implements a less-than-best-effort transport service. In this paper we present rLEDBAT, a purely receiver-based mechanism to implement LEDBAT++ for TCP. rLEDBAT enables a receiver to select some incoming traffic as less-than-best-effort, managing the capacity of the downlink. We describe the different mechanisms composing rLEDBAT that enable the execution of the LEDBAT++ congestion control algorithm at the receiver. We have implemented and experimentally tested rLEDBAT. We validate that the mechanisms incorporated by rLEDBAT at the receiver are indeed effective to implement a less-than-best-effort transport service at the receiver, as it performs similarly to the original sender-based LEDBAT++.

LEDBAT++ is the evolution of LEDBAT, a congestion control algorithm originally designed to provide less-than-best-effort transport on the Internet. LEDBAT++ aims to address a number of shortcomings present in LEDBAT, including late-comer advantage, latency drift, competition on equal grounds with best effort traffic in the presence of small buffers and difficulties experienced while measuring the variations on the delay. In this paper, we perform an experimental evaluation of LEDBAT++ using the Windows Server’s LEDBAT++ implementation. We find that while LEDBAT++ overcomes all the limitations identified in LEDBAT, the change introduced in LEDBAT++ to do so results in a performance penalty that prevents LEDBAT++ flows to seize all the available capacity when there is no competing traffic. We propose two simple modifications to the LEDBAT++ algorithm that would address the identified issues and reduce the penalty.

We present a study of the interaction between two recent congestion control algorithms, LEDBAT++ and BBR. LEDBAT++ is a less-than-best-effort congestion control mechanism and BBR is model-based congestion control algorithm that aims to implement a best-effort traffic class. Naturally, the expectation is that LEDBAT++ yields in front of BBR. While we observe that this is the case in networks with moderate size buffer and when the base RTT is larger than 60 ms, we find that in other common conditions (base RTT smaller than 60 ms and moderate size network buffers), the interaction between LEDBAT++ and BBR is not as expected, as LEDBAT++ fails to yield in front of BBR. The observed behavior is different for the two existent versions of BBR. Specifically, LEDBAT++ competes in equal grounds against BBRv1, and BBRv2 yields in front of LEDBAT++. The BBRv2/LEDBAT++ interaction is exactly the opposite to the desired outcome. We present a number of experiments in diverse conditions that support our conclusions.

Route leak events have historically caused many wide-scale disruptions on the Internet. Leaks are particularly hard to detect because they most frequently involve routes with legitimate origin announced through legitimate paths that are propagated beyond their legitimate scope. In this paper we present ASIRIA, a mechanism for detecting and avoiding leaked routes and protecting against leakage events that uses AS relationship information inferred from the Internet Routing Registries. By relying on existing information, ASIRIA provides immediate benefits to early adopters. In particular, we consider the deployment of ASIRIA to detect leaks caused by over 300 ASes and we show that it can detect over 99% of the leakage events generated by a customer or a peer solely using currently available information in 90% of the cases.

In this paper we present a large measurement study of the impact on the performance of the adoption of HTTPS as a transport for the DNS protocol (DoH) with public resolvers compared to the existent approach of using non-encrypted transport of DNS queries with the resolver services locally provided by ISPs. Using on web-ads as the mean to execute our tests, we perform over 42 million measurements from more than 4 million vantage points distributed in 32 countries and served by over 2,500 ISPs. We find that, the median resolution time increased 17 ms when using DoH with Cloudflare, 41 ms when using DoH with Quad9, 68 ms when using DoH with Google and 170 ms when using DoH with DNS.SB, compared to using Do53 with the local resolver for a non-cached name. We find similar increases even when using caching. The results presented in the paper contribute to the ongoing discussion of the tradeoffs involved in the combined adoption of public resolvers and DoH.

BGP reconvergence events involving a large number of prefixes may result in the loss of large amounts of traffic. Based on the observation that a very small number of prefixes carries the vast majority of traffic, we propose Power Prefixes Prioritization (PPP) to ensure the routes of these popular BGP prefixes converge first. By doing so, we significantly reduce the amount of traffic lost during reconvergence events. To achieve this, PPP obtains an ordered list of popular prefixes through traffic inspection, and configures the resulting prefix rank in the BGP routers to prioritize the processing and advertisement of BGP routes. We model the benefits of PPP over traditional BGP processing in terms of traffic loss for both generic and a Zipf traffic distribution, and we consider the impact of sampling in the process of obtaining the prefix rank. Applying the mechanism to real traffic traces obtained from WIDE, we show that PPP reduces the amount of traffic lost by an order of magnitude, even when we configure it to use conservative sampling rates. We prototype our proposal in Quagga to show the feasibility of its implementation, and we observe similar traffic loss reduction. PPP can be deployed incrementally, as it is implemented purely as a change in the router-internal BGP processing behavior.

Dynamic partial reconfiguration significantly reduces reconfiguration times when offloading a partial design. However, there are occasions when fine-tuning a circuit would greatly benefit from quicker reconfiguration times. To that end, authors present an automated tool and runtime support to reconfigure LUT-based multiplexers and constants. In contrast to conventional multiplexers and constants, it is possible to modify these components without having a direct communication with the static system.

Internet is growing rapidly in the developing world now. Stringent budget constraints give rise to networks with tree topology, leaving lesser room to apply energy savings methods proposed in the last decade as there are no redundant links or nodes in the network. In this paper, we propose SLEEPWELL, design of energy efficient network topology using energy aware networking devices. We divide users according to their profile of network usage and users with similar profile are all connected to one switch to allow the switch to sleep. We evaluate our framework using real data collected from the Local Area Network of IIT Kanpur (an Indian university) having more than ten thousand network users. Results show that even in a tree topology, SLEEPWELL achieves substantial energy gains, up to 22% using energy-efficient hardware, without compromising performance. We also show that dividing users in just two profiles accounts for more than 90% of the total energy saved using SLEEPWELL. We also evaluate the overheads occurred and show that the extra cost incurred can be recovered within two years for most of the practical scenarios.

Energy efficient networking has gained momentum in past one decade due to the Internet's ever increasing share in world's total energy consumption. It is crucial to study and reduce the energy consumption of tasks that are very heavily used in the Internet. In this paper, we focus on one such task, file distribution, and study its energy efficiency. We prove lower bounds on energy consumption for different scenarios that are relevant in today's Internet and design schemes that achieve the lower bounds when all the hosts have equal upload and equal download capacities. We evaluate our theoretical results numerically to generalized scenarios as well. We show that our schemes of file distribution can save as much as 50% energy compared to other energy efficient P2P methods proposed in the literature.

In this paper we propose Fast Path Ethernet, an evolution of the transparent bridges learning mechanisms to increase infrastructure utilization for campus and datacenter networks in a simple way. Fast Path Ethernet Switches reuse standard ARP Request and Reply packets to set up fast on-demand paths between hosts. This architecture uses the standard Ethernet frame format, so it is fully transparent to hosts and compatible with 802.1D bridging in core-island mode. A proof of concept has been implemented in Linux. Preliminary simulations in metropolitan and campus network topologies show superior performance to spanning tree and even to shortest path forwarding, at a fraction of the their complexity.

This paper characterizes analytically the performance of REAchability Protocol (REAP), a network layer end-to-end recovery protocol for IPv6. REAP was developed by the IETF SHIM6 Working Group as part of its multihoming solution. The behavior of REAP is governed by a small number of parameters: three timers, a simple characterization of the application traffic, and the communication delay. The key figure of merit of REAP performance is the time to recover from a path failure as seen by the upper layers, figure that cannot be trivially obtained, despite the apparent simplicity of this reachability protocol. In this paper we provide upper bounds for the recovery time of REAP for different deployment scenarios, applying these analytical results to two interesting case studies, TCP and VoIP traffic.

We present an architecture for the provision of network layer privacy based on the SHIM6 multihoming protocol. In its basic form, the architecture prevents on-path eavesdroppers from using SHIM6 network layer information to correlate packets that belong to the same communication but use different locators. To achieve this, several extensions to the SHIM6 protocol and to the HBA (Hash Based Addresses) addressing model are defined. On its full-featured mode of operation, hosts can vary dynamically the addresses of the packets of on-going communications. Single-homed hosts can adopt the SHIM6 protocol with the privacy enhancements to benefit from this protection against information collectors.

It is expected that IPv6 multihomed sites will obtain as many global prefixes as direct providers they have, so traffic engineering techniques currently used in IPv4 multihomed sites is no longer suitable. However, traffic engineering is required for several reasons, and in particular, for being able to properly support multimedia communications. In this paper we present a framework for traffic engineering in IPv6 multihomed sites with multiple global prefixes. Within this framework, we have included several tools such as DNS record manipulation and proper configuration of the policy table defined in RFC 3484. To provide automation in the management of traffic engineering, we analyzed the usage of two mechanisms to configure the policy table.