This paper studies what mechanisms a network must have to offer a very low-latency service to applications (featuring a maximum end-to-end packet delay). We assume very concrete requirements, not seen in the literature, that raise the challenge level: i) applications might be unresponsive to traffic warnings from the network; and ii) applications do not inform or require any network resources, exactly as the Internet works today (i.e., there is no admission control procedures). We present an architecture/algorithm with a minimum of state information and good scalability properties. Obviously, it is not applicable to the wide Internet. Even more, the architecture is not TCP-friendly (because control loops must be shorter than the Round Trip Time (RTT) magnitudes and oscillations, and packet losses are higher). Instead, it is appropriate to an end-to-end slice network based on a virtualization of the physical network with independent queues and line bandwidths. It is designed for interactive applications and for certain real-time ones. We use plain backpressure control supported by cooperation amongst the routers to isolate offending traffic. We are particularly concerned in situations of very high load, as they will be very common in the future. One objective is to reach a predictable network behaviour that in the limit (heavy network overload) is maintained, contrary to the current Internet. In the future, new pace-based congestion control algorithms for applications can be designed to take the most out of this type of network.

Software defined networks (SDNs) reduce the complexity of implementing multi-commodity flow (MCF) optimization problems and several optimization based SDN Traffic Engineering (TE) schemes have been proposed. This approach has several limitations like the need to know or infer bandwidth demands, the number of forwarding rules needed to implement the solution and the cost of the control-loop communication between controller and switches. This usually results in systems that do not scale and need to adapt the problem formulation to an approximation of the optimal solution. In this work we propose an approach for TE in SDNs without optimization and no need for traffic demand knowledge. We propose a logically hierarchical controller architecture built on the premise of low rule space and sparse controller communication where TE is performed in two actions: the setting of stable paths in the core and the distribution of traffic among them in the access. Different algorithms can be used for these purposes. We start by studying the performance of using a shortest path algorithm for the calculation of the stable core paths and a concept from social networks analysis called betweenness centrality to measure congestion risk and support the traffic distribution decisions in the access.1

This paper presents a scalable solution for guaranteeing bandwidth to interactions/communications assuming no information is provided by the applications. The objective is to define and assess what the network can do with so little information both in terms of guaranteeing quality and preventing being throttled by overloads. The suggestion of having applications indicating their needs did not materialize in the past and it is not credible to happen in the future. Applications tend to adapt to the conditions of the network (instead of requiring resources), and can drive the network to a halt. With no explicit information, the network must firstly protect itself from being driven into an uncontrolled saturation situation with unacceptable delays, and secondly devise what kind of guarantees can still be provided to applications. In this paper, if interactions live enough time, they become flows without any disruption in traffic or initial delays for flow establishment. Traffic is aggregated in the core and all flows access the network up to a maximum rate that is calculated when they begin. When the network becomes loaded, traffic that was granted guarantees is protected. New traffic is never blocked from accessing the network and may even get guarantees.

Software Defined Networks (SDNs) provides a separation between the control plane and the forwarding plane of networks. The software implementation of the control plane and the built in data collection mechanisms of the OpenFlow protocol promise to be excellent tools to implement Machine Learning (ML) network control applications. A first step in that direction is to understand the type of data that can be collected in SDNs and how information can be learned from that data. In this work we describe a simple architecture deployed in an enterprise network that gathers traffic data using the OpenFlow protocol. We present the data-sets that can be obtained and show how several ML techniques can be applied to it for traffic classification. The results indicate that high accuracy classification can be obtained with the data-sets using supervised learning.

Multipath routing can provide a better solution for several of the current BGP's problems in the inter-domain scenario by allowing a cleaner separation between routing and traffic engineering and better failure containment. The separation between routing control logic and data forwarding, and the use of software defined networking principles open space to easier deployments of new routing protocols at the inter-domain level. This paper explores the potential of using a multipath policy based inter-domain routing algorithm based on an algebraic model. The algorithm establishes multiple paths to a destination following a policy-based architecture and works with simple destination-based hop-by-hop forwarding. We focus on the definition of algebraic models for the links, their composition and the resulting path's ranking. Our algorithm addresses the most common situations in the Internet and we include others not so common as backup links and sibling relations. We prove the correctness of the algorithm under conditions that open several implementation options and describe how it can use multipath for failure containment, using the extra paths to make each failure less important. The paper finishes with the presentation of some experimental results that illustrate the algorithm capabilities.

Ethernet is the Layer 2 infrastructure of choice for modern data center networks and service provider metro networks. These networks need to scale to a high number of end hosts and bridges, provide isolation between tenants and support host mobility with an efficient bandwidth use and reduced operating costs. Ethernet uses controlled flooding of frames and the spanning tree protocol, falling short on demands like scalability and full usage of the topology. Solutions at Layer 3 solve some of the problems but they are harder to configure, more expensive and pose difficulties to host mobility. Recent efforts try to bring some of the advantages of L3 to L2 by using equal cost multipath (ECMP) control planes to populate the bridges forwarding tables. In this paper we go beyond and propose a non-ECMP solution where the paths are selected by policy. This means that very different paths (in terms of the physical topology) can be considered equal. It provides an extra layer of flexibility in traffic distribution with a high utilization of the network in a single L2 Ethernet domain. Our design is based on routing algebra theory and its correct behaviour can be proved for both centralized and distributed implementations.

Algebra has made the tasks of defining routing algorithms and proving their correct operation very clear. However, current results in the modelling of multipath policy routing protocols only prove correct behaviour for models that exhibit a strict decrease in the preference of a path with every added link. This is very rigid for multipath routing. For instance it does not allow a link attribute to dominate the entire path classification or having equivalent paths with similar policy but a different number of links. This paper starts by defining the fundamental sufficient conditions for the correct operation of a multipath policy routing protocol with independent destination based hop-by-hop forwarding. We then study the sufficient properties that the protocol model should exhibit so that those conditions are met. We prove that correct operation is possible without the strict increase in preference, provided that the policies applied to links forming circuits in the network graph are restricted. We derive two conditions on the policies applied in the circuits of a network graph that assure correct behaviour and can serve as a tool in the design process.

Traffic Engineering (TE) aims at distributing traffic through routes to suit a specific goal. BGP addresses TE in a very specific way: it is a single path system, and mixes traffic control and route dissemination by letting attributes influence the choice of the best path. Several techniques have been proposed but due to BGP characteristics they have high complexity requirements such as the need for coordination and information about the network that is not locally available. and, in most cases, the use of tunnelling. In a previous work we proposed an inter-domain routing architecture DTIA (Dynamic Topological Information Architecture) that builds a robust multipath routing system based on the inter-AS relationships and routing policies. DTIA separates traffic control from route dissemination and seems to have less complex requirements to provide TE. The aim of this paper is to assess the ability of such a system to provide TE maintaining the complexity overhead to a minimum. We propose a simple feedback protocol to avoid congestion and achieve better traffic distribution at inter-domain level using only local available traffic information and DTIA's routing information.

DTIA (Dynamic Topological Information Architecture), an inter-domain routing architecture, assumed the existence of "regions of domains" to offer a scalable solution for the Internet. The way regions can be connected and how traffic flow amongst them is not a trivial aspect and can jeopardize the solution proposed for intra-region traffic. This paper covers the inter-region case showing that it can be performed by using a smaller set of valid paths than for the intra-region case in the source region, but still featuring multipath to the destination and taking advantage of multi-homing. The routing is based on a monotone algebra to converge without the need to change the IP headers or introduce additional signaling or state in the network. The overall system complies with the current business model of the Internet, and it also features one tendency that has been increasingly used: the fact that Autonomous Systems (ASes) are each time more connected with direct (or semi-direct) links instead of in a pure hierarchical structure up to tier-1 ASes.

This paper proposes an architecture for inter-domain routing, called DTIA - Dynamic Topological Information Architecture. DTIA separates the issues of reachability and routing, and this paper addresses the first one. One major requirement has been not to change IP packets and the commercial relations in the Internet. DTIA is based on the knowledge of a static network formed by the Autonomous Systems (AS) and an algorithm to manage link failures. We use the concept of a region as a mechanism to sustain scale. DTIA supports the most important functionalities of BGP: some of them are built in and others can be implemented on top of the reachability level or the routing level. The main concerns we aim to solve are taking advantage of multihoming, increase the robustness in terms of convergence, reduce the churn rate and range of routing events, and due to forwarding packets by AS identifiers and topologic links (as opposed to prefix policy defined) reduce the growth of the routing table.

This paper proposes an architecture for inter-domain routing, called DTIA -- Dynamic Topological Information Architecture. It is designed to address current limitations of BGP like: scalability of the routing tables, multihoming problems, churn rate, range of routing events and policy coordination. Still it supports the most important functionalities of BGP, and maintains the current Internet routing distribution model based on Inter-AS business relationships. The architecture is based on the knowledge of a static network graph formed by the Autonomous Systems (AS) and an algorithm to manage link failures. We use the concept of a region as a mechanism to sustain scale. New possible features like multi-path routing and enhanced Traffic Engineering capabilities might benefit with DTIA's usage.

This paper presents the Split and Merge method, an approach to provide secure communication over the Internet using software agents. Instead of relying exclusively on the traditional approaches of ciphering the message at the source and then sending it over a communication channel, the Split and Merge method focuses on denying access to the message itself. This is achieved by splitting the message in parts and sending them to the destination through different routes, in an ad-hoc network of agents. The method is detailed in the paper and the current implementation using multi agent systems is presented. The solutions found to deal with error detection and correction (loss of message fragments, message integrity and node misbehavior) are then discussed and further research directions are presented.