Many emerging distributed applications, including big data analytics, generate a number of flows that concurrently transport data across data center networks. To improve their performance, it is required to account for the behavior of a collection of flows, i.e., coflows, rather than individual. State-of-the-art solutions allow for a near-optimal completion time by continuously reordering the unfinished coflows at the end-host, using network priorities. This paper shows that dynamically changing flow priorities at the end host, without taking into account in-flight packets, can cause high-degrees of packet re-ordering, thus imposing pressure on the congestion control and potentially harming network performance in the presence of switches with shallow buffers. We present pCoflow, a new solution that integrates end-host based coflow ordering with in-network scheduling based on packet history. Our evaluation shows that pCoflow improves in CCT upon state-of-the-art solutions by up to 34% for varying load.

Traffic management in data centers is paramount to improving network and application performance, thereby improving the quality of service by reducing network congestion, packet loss, and latency. However, the deployment and configuration of traffic management techniques are challenging due to diverse data-center traffic characteristics, large data center topologies, and the interplay of different protocols at the routing, transport, and link layer. Software-Defined Networking (SDN) emerges as a new paradigm towards a centralized network configuration and traffic management by decoupling the control plane from forwarding devices. Despite its holistic view of the network, data centers are commonly interconnected by traditional networks that use standard routing protocols. It is therefore essential to achieve interoperability with legacy systems, end-to-end traffic management, and to avoid the cumbersome, time-consuming, and error-prone configuration process of data-center edge network devices.

In this thesis, we aim to improve traffic management and its configuration for software-defined data center networks. To achieve this objective, we provide novel approaches that enhance the control plane as well as leverage novel concepts of data plane programmability.

At the control plane, we first propose different mechanisms that enable the fast restoration of network connectivity after a virtual machine migration. Second, we suggest a network management automation framework that extends layer 2 connectivity to the tenants' services hosted across geo-distributed data centers. Moreover, we provide high-level policy-based mechanisms that make network configuration and traffic management simpler for data-center operators. At the data plane, we develop MP-HULA that load-balances multipath connections across least-congested paths. MP-HULA leverages advanced data plane mechanisms to rank multiple paths according to congestion metrics and uses that information for fine-grained load-balancing decisions considering transport layer information. To improve flowlet-based load-balancing decisions, we propose FlowDyn, which efficiently estimates round-trip time using programmable telemetry data. Finally, we present pCoflow, an in-network support mechanism that uses advanced programmable scheduling primitives to effectively avoid reordering for data-parallel applications even when there are flow priority changes due to global coflow scheduling updates.

Traffic management in data centers is paramount to improving network and application performance, thereby improving the quality of service by reducing network congestion, packet loss, and latency. However, the deployment and configuration of traffic management techniques are challenging due to diverse data center traffic characteristics, large data center topologies, and the interplay of different protocols at the routing, transport, and link layer. Software-Defined Networking (SDN) emerges as a new paradigm towards a centralized network configuration and traffic management by decoupling the control plane from forwarding devices.

In this thesis, we aim to improve traffic management and its configuration for software-defined data center networks by providing novel approaches that enhance the control plane as well as leveraging novel concepts of data plane programmability. At the control plane, we provide solutions such as high-level based policy framework, automation of layer 2 services, and fast restoration of network connectivity for virtual machine migration, that make network configuration and traffic management simpler for data-center operators. At the data plane, we propose approaches to improve load balancing in data centers and an in-network support mechanism to avoid reordering for data-parallel applications even when there are flow priority changes due to global coflow scheduling updates.

Data center networks offer multiple disjoint paths between Top-of-Rack (ToR) switches to connect server racks providing large bisection bandwidth. An effective load-balancing mechanism is required in order to fully utilize the available capacity of the multiple paths. While packet-based loadbalancing can achieve high utilization, it suffers from reordering. Flow-based load-balancing such as equal-cost multipath routing (ECMP) spreads traffic uniformly across multiple paths leading to frequent hash collisions and suboptimal performance. Finally, flowlet based load-balancing such as CONGA or HULA splits flows into smaller units, which are sent on different paths. Most flowlet based load-balancing schemes depend on a proper static setting of the flowlet gap, which decides when new flowlets are detected. While a too small gap may lead to reordering, a too large gap results in missed load-balancing opportunities. In this paper,weproposeFlowDyn,whichdynamicallyadaptstheflowlet gap to increase the efficiency of the load-balancing schemes while avoiding the reordering problem. Using programmable data planes, FlowDyn uses active probes together with telemetry informationtotrackpathlatencybetweendifferentToRswitches. FlowDyn calculates dynamically a suitable flowlet gap that can be used for flowlet based load-balancing mechanism. We evaluate FlowDyn extensively in simulation, showing that it achieves 3.19 times smaller flow completion time at 10% load and 1.16x at 90% load.

Datacenter networks ofer a large degree of multipath in order to provide large bisectional bandwidth. The end-to-end performance is determined by the load-balancing strategy which needs to be designed to efectively manage congestion. Consequently, congestion aware load-balancing strategies such as CONGA or HULA have been designed. Recently, more and more applications that are hosted on cloud servers use multipath transport protocols such as MPTCP. However, in the presence of MPTCP, existing load-balancing schemes including ECMP, HULA or CONGA may lead to suboptimal forwarding decisions where multiple MPTCP subfows of one connection are pinned on the same bottleneck link. In this paper, we present MP-HULA, a transport layer multi-path aware load-balancing scheme using Programmable Data Planes. First, instead of tracking congestion information for the best path towards the destination, each MP-HULA switch tracks congestion information for the best-k paths to a destination through the neighbor switches. Second, we design MP-HULA using Programmable Data Planes, where each leaf switch can identify, using P4, which MPTCP subfow belongs to which connection. MP-HULA then load-balances diferent MPTCP subfows of a MPTCP connection on diferent next hops considering congestion state while aggregating bandwidth. Our evaluation shows that MP-HULA with MPTCP outperforms HULA in average flow completion time (2.1x at 50% load, 1.7x at 80% load).

Both Big Data and SDN are described in detail in previous chapters. This chapter investigates how SDN architecture can leverage its unique features to mitigate the challenges of Big Data volume. Accordingly, first, we provide an overview of Big Data volume, its effects on the underlying network, and mention some potential SDN solutions to address the corresponding challenges. Second, we elaborate more on the network-monitoring, traffic-engineering, and fault-tolerant mechanisms which we believe they may help to address the challenges of Big Data volume. Finally, this chapter is concluded with some open issues.

This chapter begins by explaining the main SDN concepts with the focus on a SDN controller. It presents the most important aspects to consider when we desire to go from traditional network to a SDN networks. We present an in-depth analysis of the most commonly used and modern SDN controllers and analyse the main features, capabilities and requirements of one of the presented controllers. OpenFlow is the standard leading in the market allowing the management of the forwarding plane devices such as routers or switches. While there are other standards with the same aim, OpenFlow has secured a position in the market and has been expanded rapidly. Therefore, an analysis is presented on a different OpenFlow compatible device for the implementation of an SDN network. This study encompasses both software and hardware solutions along with the scope of implementation or use of these devices. This chapter ends up presenting a description of OpenFlow protocol alternatives, a more detailed description of OpenFlow and its components and other wellknown southbound protocols involved for the management and configuration of the devices.

SDN aims to facilitate the management of increasingly complex, dynamic network environments and optimize the use of the resources available therein with minimal operator intervention. To this end, SDN controllers maintain a global view of the network topology and its state. However, the extraction of information about network flows and other network metrics remains a non-trivial challenge. Network applications exhibit a wide range of properties, posing diverse, often conflicting, demands towards the network. As these requirements are typically not known, controllers must rely on error-prone heuristics to extract them. In this work, we develop a framework which allows applications deployed in an SDN environment to explicitly express their requirements to the network. Conversely, it allows network controllers to deploy policies on end-hosts and to supply applications with information about network paths, salient servers and other relevant metrics. The proposed approach opens the door for fine grained, application-aware resource optimization strategies in SDNs

Layer 2 Virtual Private Network (L2VPN) is widely deployed in both service provider networks and enterprises. However, legacy L2VPN solutions have scalability limitations in the context of Data Center (DC) interconnection and networking which require new approaches that address the requirements of service providers for virtual private cloud services. Recently, Ethernet VPN (EVPN) has been proposed to address many of those concerns and vendors started to deploy EVPN based solutions in DC edge routers. However, manual configuration leads to a time-consuming, error-prone configuration and high operational costs. Automating the EVPN deployment from cloud platforms such as OpenStack enhances both the deployment and flexibility of EVPN Instances (EVIs). This paper proposes a Software Defined Network (SDN) based framework that automates the EVPN deployment and management inside SDN-based DCs using OpenStack and OpenDaylight (ODL). We implemented and extended several modules inside ODL controller to manage and interact with EVIs and an interface to OpenStack that allows the deployment and configuration of EVIs. We conclude with scalability analysis of our solution.

OpenStack has been widely acknowledged to be one of the most important open source cloud platforms. In order to perform experimentally driven research in the area of cloud and cloud networking, there is however a big gap, because most researchers do not have access to a large cloud deployment and cannot change networking or compute infrastructure in order to test their algorithms and protocols on a large-scale. We developed OpenStackEmu, which is to the best of our knowledge the first attempt that combines OpenStack infrastructure with a Software Defined Networking (SDN) based controller such as OpenDaylight and a large-scale network emulator CORE (Common Open Research Emulator). The OpenStack compute and control nodes are connected to the CORE emulation server using TUN/TAP interfaces that inject the control (e.g. for VM migration) and data (VM-to-VM traffic) packets into a customizable network topology that is emulated using configurable Open vSwitches using CORE emulator. Experimenters can define e.g. fat-tree or distributed data center topologies and study the behavior of real VMs and services in those VMs under different background loads and SDN routing policies. We integrated the data center traffic generator DCT2Gen that allows to generate realistic background traffic based on traces from real data centers. Experimenters can study the performance impact of different VM migration strategies or different routing and load balancing schemes on real VM and application performance using different emulated topologies. We believe that OpenStackEmu is an important tool for both the SDN and OpenStack community in order to evaluate the performance of novel algorithms and protocols in the area of cloud networking.

The Ethernet VPN (EVPN) technology has emerged as a key solution for the interconnection of geo-distributed Data Centers (DCs) over provider-managed MPLS networks. Such interconnections need to satisfy service-level agreements, which can be achieved by enforcing Traffic Engineering (TE) policies. However, deploying an effective TE policy is challenging and complex. This stems from the fact that network administrators should have a detailed insight into the network status and protocol specifics. Software-Defined Networking (SDN) may facilitate both the policy definition and deployment based on its comprehensive network view and existing integration with DC management systems, such as OpenStack. This paper presents an SDN-based framework for policy-driven DC interconnections that are built around EVPN. The framework is designed to translate routing and other TE policies, which are defined for EVPN instances, into appropriate low-level network actions to meet the policy goals. A generic programming interface allows an SDN controller to load different TE strategies so as to implement the policy, without the need to hard-code it. Moreover, our evaluations illustrate how clients might benefit from specific TE strategies and what is their impact on network performance