Simone Ferlin

Experimental evaluation of multipath TCP schedulers

Today many end hosts are equipped with multiple interfaces. These interfaces can be utilized simultaneously by multipath protocols to pool resources of the links in an efficient way while also providing resilience to eventual link failures. However how to schedule the data segments over multiple links is a challenging problem, and highly influences the performance of multipath protocols. In this paper, we focus on different schedulers for Multipath TCP. We first design and implement a generic modular scheduler framework that enables testing of different schedulers for Multipath TCP. We then use this framework to do an in-depth analysis of different schedulers by running emulated and real-world experiments on a testbed. We consider bulk data transfer as well as application limited traffic and identify metrics to quantify the schedulers performance. Our results shed light on how scheduling decisions can help to improve multipath transfer.

Multi-path transport over heterogeneous wireless networks: Does it really pay off?

Multi-path transfer protocols such as Concurrent Multi-Path Transfer for SCTP and Multi-Path TCP (MPTCP), are becoming increasingly popular, due to widespread deployment of smartphones with multi-homing support. Although the idea of using multiple interfaces simultaneously to improve application throughput is tempting, does transmission over multiple interfaces always provide benefits especially in realistic setup? In this paper, we first show that multi-path transfer might actually have a negative impact in real-world scenarios with mobile broadband and wireless LAN networks. We then introduce our Dynamic Relative Path Scoring (DRePaS) algorithm that continuously evaluates the contribution of paths to the overall performance and dynamically influences the scheduling decisions to make best use of the paths for the overall system performance. We show that DRePaS outperforms the current MPTCP implementation in terms of throughput and application delay, especially when the links are heterogeneous.

BLEST: Blocking estimation-based MPTCP scheduler for heterogeneous networks

With the widespread availability of multi-homed devices, multipath transport protocols such as MPTCP are becoming increasingly relevant to support better use of multiple connectivity through capacity aggregation and seamless failover. However, capacity aggregation over heterogeneous paths, such as offered by cellular and Wi-Fi networks, is problematic. It causes packet reordering leading to head-of-line (HoL) blocking at the receiver, increased end-to-end delays and lower application goodput. MPTCP tackles this issue by penalising the use of longer paths, and increasing buffer sizes. This, however, results in suboptimal resource usage. In this paper, we first evaluate and compare the performance of default MPTCP and alternative state-of-the-art schedulers, all implemented in the Linux kernel, for a range of traffic patterns and network environments. This allows us to identify shortcomings of various approaches. We then propose a send-window BLocking ESTimation scheduler, BLEST, which aims to minimise HoL-blocking in heterogeneous networks, thereby increasing the potential for capacity aggregation by reducing the number of spurious retransmissions. The resulting scheduler allows an increase by 12% in application goodput with bulk traffic while reducing unnecessary retransmissions by 80% as compared to default MPTCP and other schedulers.

Measuring the QoS Characteristics of Operational 3G Mobile Broadband Networks

Today, many smart phones and tablets have multiple interfaces (i.e. WLAN and 3G). These multiple interfaces can be utilized simultaneously by a multi-path transport protocol to provide bandwidth aggregation or reliability. However, in order to design efficient multi-path scheduling and congestion control strategies, it is crucial to understand the behaviour and properties of the underlying paths first. WLAN links have already been studied extensively in the literature. Therefore, in this paper, we focus on Mobile Broadband (MBB) networks that are in use today. We utilized noun Nor Net Edge nodes that are connected to up to five different 3G ISPs (UMTS and CDMA2000), hence, providing a realistic view on the QoS characteristics that are experienced by end-users of these MBB networks. We present QoS characteristics (e.g. Bandwidth, delay and loss) and discuss our observations. Our results shed light on what a multi-path transport endpoint has to expect - and to efficiently cope with - when using todays MBB networks as transport paths.

An Early Look at Multipath TCP Deployment in the Wild

Multipath TCP was standardised in 2013 at IETF. It promises better use of network resources of multi-homed devices for capacity aggregation or seamless fail-over capabilities. The uptake has however been rather slow. Some operating systems support MPTCP out of the box, but little is known about their deployment on the server side. We built a scanning infrastructure to search for MPTCP-capable hosts on the Internet. In this study, we used the top-1M Alexa servers to test the platform and gain insight about server support. We find that less than 0.1% of Alexa unique domains and IP addresses currently support MPTCP. Their geographic distribution is quite different from that of clients reported in other studies, with the majority of domains being in China. We also find that very few IPs actually expose multi-homing information, suggesting that these early deployments aim at providing reliability rather than capacity aggregation. We also identify some deployment issues.

Leveraging the IPv4/IPv6 identity duality by using multi-path transport

With the 20th anniversary of IPv6 nearing quickly, a growing number of Internet service providers (ISPs) now offer their customers both IPv6 and IPv4 connectivity. This makes multi-homing with IPv4 and IPv6 increasingly common even with just a single ISP connection. Furthermore, the growing popularity of multi-path transport, especially Multi-Path TCP (MPTCP) that is the extension of the well-known Transmission Control Protocol (TCP), leads to the question of whether this identity duality can be utilized for improving application performance in addition to providing resilience. In this paper, we first investigate the AS-level congruency of IPv4 and IPv6 paths in the Internet. We find that more than 60% of the current IPv4 and IPv6 AS-paths are non-congruent at the AS-level, which motivates us to explore how MPTCP can utilize the IPv4/IPv6 identity duality to improve data transfer performance. Our results show that MPTCP, even with a single dual-stack Internet connection, can significantly improve the end-to-end performance when the underlying paths are non-congruent. The extent of the improvement can reach up to the aggregate of the IPv4 and IPv6 bandwidths.

Revisiting congestion control for multipath TCP with shared bottleneck detection

Multipath TCP (MPTCP) enables the simultaneous use of multiple links for bandwidth aggregation, better resource utilization and improved reliability. Its coupled congestion control intends to reap the increased bandwidth of multiple links, while avoiding being more aggressive than regular TCP flows on every used link. We argue that this leads to a very conservative behavior when paths do not share a bottleneck. Therefore, in this paper, we first quantify the penalty of the coupled congestion control for links that do not share a bottleneck. Then, in order to overcome this penalty, we design and implement a practical shared bottleneck detection (SBD) algorithm for MPTCP, namely MPTCP-SBD. Through extensive emulations, we show that MPTCP-SBD outperforms all currently deployed MPTCP coupled congestion controls by accurately detecting bottlenecks. For the non-shared bottleneck scenario, we observe throughput gains of up to 40% with two subflows and the gains increase significantly as the number of subflows increase, reaching more than 100% for five subflows. Furthermore, for the shared bottleneck scenario, we show that MPTCP-SBD remains fair to TCP. We complement the emulation results with real-network experiments justifying its safeness for deployment.

Practical passive shared bottleneck detection using shape summary statistics

Practical shared bottleneck detection has proved to be a difficult problem. We present a novel passive approach using efficient estimates of time and frequency domain summary statistics. The approach is not CPU nor network intensive, and has numerous potential applications in the Internet. Simulations and tests over the Internet and 3G cellular network show its efficacy in grouping flows correctly.

LISA: A linked slow-start algorithm for MPTCP

One of the main goals of multipath TCP (MPTCP) is to achieve higher throughput than regular TCP by utilizing multiple paths simultaneously. When these paths share a common bottleneck, MPTCP tries not to be more aggressive than a regular TCP flow. This is achieved by MPTCPs coupled congestion control mechanism that couples the increase factor of MPTCPs subflows in congestion avoidance. However, slow-start remains unchanged and behaves uncoupled for each subflow, affecting MPTCP and concurrent traffic at the bottleneck. We propose LISA, a simple algorithm for coupling MPTCP subflows in slow-start, and investigate the trade-off that this coupling entails. Our evaluations show that coupling in slow-start not only provides gains for MPTCP but also for a concurrent TCP at the bottleneck.