• Keine Ergebnisse gefunden

apply it to solve optimization problems for eciently using the communication infras-tructure in smart factories.

1.5 Structure of this Thesis

This thesis is structured as follows. In Chapter 2, we present a brief background of the networking technologies that are used in this thesis. In particular, we introduce the extensions to the IEEE 802.1Q standards relevant for handling scheduled trac like time-triggered data streams. We also present a brief summary of OpenFlow, a popular SDN southbound protocol, used for implementing the routing algorithms in this thesis.

Chapter 3 presents our unied network model, specically our assumptions pertaining to the network data plane and the network control plane, along with our model of time-triggered data streams.

In Chapter 4, we introduce the mapping of the scheduling problem in the networks compliant with the IEEE 802.1Qbv standard to the No-wait Job-shop Scheduling Prob-lem (NW-JSP) and present heuristics based on Tabu-search for eciently solving this problem. This chapter also presents the usage of Intel's Data Plane Development Kit (DPDK) for precise adherence to the computed schedules by the end systems.

Chapter 5 introduces the routing problem, i.e., the impact of routing time-triggered data streams on the computed schedules, for the networks compliant with the IEEE 802.1Qbv enhancements. It also presents ILP based formulations for computing routes for time-triggered data streams in such networks.

The architecture of TSSDN and the static scheduling and routing problem in networks where the schedules are enforced at the end systems alone is presented in Chapter 6.

For this problem, we not only present an ILP formulation that computes the optimal schedules and routes, but also two heuristic solutions by restricting the number of paths over which time-triggered streams can be routed.

The dynamic scheduling problem in TSSDN is presented in Chapter 7. We present optimized scheduling algorithms that can compute the incremental schedules and routes for time-triggered data streams in realistic scenarios under a second.

In Chapter 8, we introduce the routing problem for time-triggered data streams in networks where the network participants cannot enforce schedules due to lack of syn-chronized clocks. The chapter also presents ecient algorithms based on meta-heuristic approaches to compute these routes for time-triggered data streams.

We conclude the thesis in Chapter 9 with a brief discussion of our contributions and an outlook towards future research in this direction.

1 Introduction

Summary

With the advent of Industrial Internet of Things (IIoT), there is a growing desire to have converged networks for transporting time-sensitive trac along with best-eort trac in manufacturing systems. The proliferation of Ethernet along with its rapid pace of development has made it a strong contender for this purpose. However, Ethernet being designed for providing best-eort communication services, is not suitable for handling real-time trac without further enhancements. An idea to provide bounded end-to-end latency and jitter for time-triggered data streams, for instance, a stream of samples transmitted by a sensor, in Ethernet is to schedule the traversal of such streams through the network such that the indeterministic queuing delay encountered by these streams is bounded. The computation of these schedules must take into account the capabilities of the network participants, i.e., who would be enforcing the schedules. Another major factor that aects the computation of schedules for time-triggered data streams is the paths over which these streams are routed.

This thesis mainly deals with dierent scheduling and routing problems, most of which have a high time complexity, with respect to the handling of time-triggered data streams in Ethernet networks. In particular, it classies the solution space for scheduling the time-triggered data streams into four parts based on where the computed schedules are to be enforced. This thesis provides scheduling and routing solutions for each of these dierent scenarios.

14

CHAPTER 2

BACKGROUND

This chapter mainly presents dierent networking technologies and relevant IEEE stan-dards that we refer to in this thesis for implementing routing and scheduling of time-triggered data streams.

2.1 Software-dened Networking

As already mentioned in Chapter 1, Software-dened networking (SDN) aims to im-prove exibility of computer networks. The SDN paradigm is based on two main principles, viz., the separation of the network control plane from the network data plane and the logical centralization of the control plane with a global view on the data plane. The control plane interacts and congures the data plane using standardized interfaces known as the southbound protocols. OpenFlow, from the Open Networking Foundation, is one of the most popular southbound protocols and is on the path to become the de-facto SDN southbound protocol [41].

OpenFlow is a communication protocol between the data plane elements (switches) and the network controller hosting the control plane. Using OpenFlow, the network controller can program the routing tables of the switches, thus inuencing the routes of the trac in the data plane. In this thesis, we use OpenFlow for routing time-triggered data streams based on their computed schedules and routes in the Time-sensitive Software-dened Network (TSSDN) presented in Chapter 6. Though Open-Flow is well-documented, this section briey describes the working of the protocol for the sake of completeness.

2 Background

2.1.1 OpenFlow

OpenFlow switches process received packets based on the entries in its routing tables, also known as ow tables. Each entry (cf. Table 2.1) in the ow table represents a forwarding rule consisting of match elds, instructions and additional information for packet processing. Match elds store the values for the layer 24 header elds of the network stack which must match with those of the packet being processed for the rule to apply. If the rule applies for the packet being processed, then the actions specied in the instructions eld will be executed on the packet. The possible actions range from rewriting some header elds like the source or destination MAC addresses to forwarding the packet over a certain egress port or even dropping the packet.

Additional information includes counters which store the number of packets that matched the corresponding ow table entry and were accordingly processed. The pri-ority value is used to resolve conicts when multiple ow table entries are applicable for any of the packets. The timeouts specify the amount of time for which the ow table entry is valid. The timeouts can be specied in terms of maximum amount of time for which the rule exists (hard time-out) or in terms of idle time after which the rule expires (soft time-out). OpenFlow also supports wildcarding of elds to have coarse-grained matches for the ow tables, thus enabling multiple ows to match on the same rule. It also allows chaining of actions to create a pipeline for more complex operations in the data plane.

Match Fields Instructions Counters Priority Timeouts Cookie Table 2.1: Flow table entries in an OpenFlow switch [42].

The OpenFlow protocol allows the SDN controller to connect to OpenFlow switches over a secure channel and read/write ow table entries, and thus, inuence the routing of trac through the network. The possibility of dynamically updating the ow tables brings its own set of challenges, e.g., the consistency of ow table entries across the switches in the network [43].

Recent versions of OpenFlow (since v1.4) provide transactional (all or nothing) se-mantics for updating ow table rules to avoid these consistency issues. We exploit these features for dynamically setting up time-triggered data streams in TSSDN (cf.

Chapter 7) while avoiding problems for the existing data streams.