Fabien Duchene

Exploring mobile/WiFi handover with multipath TCP

Mobile Operators see an unending growth of data traffic generated by their customers on their mobile data networks. As the operators start to have a hard time carrying all this traffic over 3G or 4G networks, offloading to WiFi is being considered. Multipath TCP (MPTCP) is an evolution of TCP that allows the simultaneous use of multiple interfaces for a single connection while still presenting a standard TCP socket API to the application. The protocol specification of Multipath TCP has foreseen the different building blocks to allow transparent handover from WiFi to 3G back and forth. In this paper we experimentally prove the feasibility of using MPTCP for mobile/WiFi handover in the current Internet. Our experiments run over real WiFi/3G networks and use our Linux kernel implementation of MPTCP that we enhanced to better support handover. We analyze MPTCPs energy consumption and handover performance in various operational modes. We find that MPTCP enables smooth handovers offering reasonable performance even for very demanding applications such as VoIP. Finally, our experiments showed that lost MPTCP control signals can adversely affect handover performance; we implement and test a simple but effective solution to this issue.

FlowBender: Flow-level Adaptive Routing for Improved Latency and Throughput in Datacenter Networks

Datacenter networks provide high path diversity for traffic between machines. Load balancing traffic across these paths is important for both, latency- and throughput-sensitive applications. The standard load balancing techniques used today obliviously hash a flow to a random path. When long flows collide on the same path, this might lead to long lasting congestion while other paths could be underutilized, degrading performance of other flows as well. Recent proposals to address this shortcoming incur significant implementation complexity at the host that would actually slow down short flows (MPTCP), depend on relatively slow centralized controllers for rerouting large congesting flows (Hedera), or require custom switch hardware, hindering near-term deployment (DeTail). We propose FlowBender, a novel technique that: (1) Load balances distributively at the granularity of flows instead of packets, avoiding excessive packet reordering. (2) Uses end-host-driven rehashing to trigger dynamic flow-to-path assignment. (3) Recovers from link failures within a Retransmit Timeout (RTO). (4) Amounts to less than 50 lines of critical kernel code and is readily deployable in commodity data centers today. (5) Is very robust and simple to tune. We evaluate FlowBender using both simulations and a real testbed implementation, and show that it improves average and tail latencies significantly compared to state of the art techniques without incurring the significant overhead and complexity of other load balancing schemes.

Are TCP extensions middlebox-proof?

Besides the traditional routers and switches, middleboxes such as NATs, firewalls, IDS or proxies have a growing importance in many networks, notably in entreprise and wireless access networks. Many of these middleboxes modify the packets that they process. For this, they to implement (a subset of) protocols like TCP. Despite the deployment of these middleboxes, TCP continues to evolve on the endhosts and little is known about the interactions between TCP extensions and the middleboxes. In this paper, we experimentally evaluate the interference between middleboxes and the Linux TCP stack. For this, we first propose MBtest, a set of Click elements that model middlebox behavior. We use it to experimentally evaluate how three TCP extensions interact with middleboxes. We also analyzes measurements of the interference between Multipath TCP and middleboxes in fifty different networks.

Leveraging eBPF for programmable network functions with IPv6 Segment Routing

With the advent of Software Defined Networks (SDN), Network Function Virtualisation (NFV) or Service Function Chaining (SFC), operators expect networks to support flexible services beyond the mere forwarding of packets. The network programmability framework which is being developed within the IETF by leveraging IPv6 Segment Routing enables the realisation of in-network functions. In this paper, we demonstrate that this vision of in-network programmability can be realised. By leveraging the eBPF support in the Linux kernel, we implement a flexible framework that allows network operators to encode their own network functions as eBPF code that is automatically executed while processing specific packets. Our lab measurements indicate that the overhead of calling such eBPF functions remains acceptable. Thanks to eBPF, operators can implement a variety of network functions. We describe the architecture of our implementation in the Linux kernel. This extension has been released with Linux 4.18. We illustrate the flexibility of our approach with three different use cases: delay measurements, hybrid networks and network discovery. Our lab measurements also indicate that the performance penalty of running eBPF network functions on Linux routers does not incur a significant overhead.

Exploring various use cases for IPv6 Segment Routing

IPv6 Segment Routing (SRv6) is a modern version of source routing that is being standardised within the IETF to address a variety of use cases in ISP, datacenter and entreprise networks. Its inclusion in recent versions of the Linux kernel enables researchers to explore and extend this new protocol. We leverage and extend the SRv6 implementation in the Linux kernel to demonstrate two very different usages of this new protocol. We first show how entreprise networks can leverage SRv6 to better control the utilisation of their infrastructure and demonstrate how DNS resolvers can act as SDN controllers. We then demonstrate how SRv6Pipes can be used to efficiently implement network functions that need to process bytestreams on top of a packet-based SRv6 network.

Making multipath TCP friendlier to load balancers and anycast

Multipath TCP is a recent TCP extension that enables the utilization of different paths for a single connection. This provides various benefits including bandwidth aggregation and fast handovers on mobiles. A Multipath TCP connection starts with a single TCP connection called subflow and other subflows are added later to increase bandwidth or support failover. One drawback of Multipath TCP is that it is not currently compatible with stateless load balancers which rely on the five-tuple for their forwarding decision. This hinders the deployment of Multipath TCP. We show that this limitation can be circumvented with a small change to the handling of the initial subflow. Clients use this subflow to discover the load-balanced server and the additional Multipath TCP subflows are terminated at a unique address associated to each physical server. With this small change, Multipath TCP becomes compatible with existing stateless load balancers. Furthermore, we show that the same approach enables anycast Multipath TCP services, a major benefit given the difficulty of deploying anycast TCP services. We implement this modification in the Linux kernel and demonstrate its benefits with several micro benchmarks.