Evaluation of Dynamic Reconfiguration Architecture in Multi-Hop Switched Ethernet Networks Mohammad Ashjaei 1 , Paulo Pedreiras 2 , Moris Behnam 1 , Luis Almeida 3 , Thomas Nolte 1 1 MRTC/M¨ alardalen University, V¨ aster˚ as, Sweden 2 DETI/IT/University of Aveiro, Aveiro, Portugal 3 IT/DEEC/University of Porto, Porto, Portugal Abstract—On-the-fly adaptability and reconfigurability are recently becoming an interest in real-time communications. To assure a continued real-time behavior, the admission control with a quality-of-service mechanism is required, that screen all adaptation and reconfiguration requests. In the context of switched Ethernet networks, the FTT-SE protocol provides adaptive real-time communication. Recently, we proposed two methods to perform the online reconfiguration in multi-hop FTT- SE architectures. However, the methods lack the experimental evaluation. In this paper, we evaluate both methods in terms of the reconfiguration time. I. I NTRODUCTION The interest of using Ethernet switches in real-time dis- tributed applications is rapidly increasing due to its features such as wide availability, low cost and high throughput. How- ever, using commercially available (COTS) switches in time critical applications may hinder the ability to provide real-time guarantees. In addition, operating conditions may change, for example triggered by changes in the environment that may lead to increased communication requirements. In turn, this calls upon adequate dynamic adaptation and reconfiguration policies that ensure continued timeliness in the communications. The limitations imposed by the simple use of COTS switches have been addressed in [1]. Some relatively old solutions are based on enhanced switches such as EtheReal [2] and the EDF Scheduled Switch [3], both using reserved channels for traffic transmission. Some other solutions made it to the market, such as PROFINET-IRT [4] and TTEthernet [5], both optimized for time-triggered operation. However, these switches are configured in ways that are not suited for dynamic real-time systems that are operating in dynamic environments. despite the performance improvements offered by using these enhanced switches, their usage result in a high cost and a lower availability compared to COTS switches. A more effective solution is to control the traffic submitted to the COTS switch avoiding queue build up and then to use adequate traffic scheduling policies. This can be achieved with a master-slave technique which is the case of the FTT-SE protocol [6]. The FTT-SE (Flexible Time-Triggered Switched Ethernet) protocol is a bandwidth-efficient master-slave protocol that handles all types of message streams including real-time peri- odic, real-time sporadic and non-real-time traffic. The protocol provides temporal isolation between the message types by defining specific reserved bandwidth for each type of message streams. Moreover, it caters for requirements of dynamic reconfiguration and adaptability. The multi-hop communication over the FTT-SE protocol was addressed in [7] and [8], where three architectures were studied. It turned out that, among those architectures, the one where several master nodes coexist to control the traffic in a group of switches performs better in terms of bandwidth utilization [8]. This architecture is called hybrid architecture. Recently, we proposed two different methods for on-line reconfiguration in the hybrid architecture [9]: centralized and distributed. However, the methods lack the experimental eval- uation. In this paper, we evaluate both methods in terms of the reconfiguration time. The paper organizes as follow. The next section describes the hybrid FTT-SE architecture. Section III presents the re- configuration methods. Section IV depicts the evaluation of the methods and Section V concludes the paper and shows future directions. II. THE HYBRID ARCHITECTURE An example of the hybrid architecture is depicted in Fig. 1. In this architecture, a group of switches along with their associated nodes that have the same parent switch form a cluster (e.g., Cluster2 in Fig. 1). The traffic within a cluster is controlled by one master node connected to the parent switch of the cluster (e.g., M2 is a master node for Cluster2). Note that, the master of the root cluster (M1 in Fig. 1) is included in its cluster since it cannot be accounted as one cluster itself. M1 M2 SW1 SW3 SW2 SW5 SW4 Cluster2 Cluster1 S1 S2 S5 S4 S3 Fig. 1. The Hybrid Architecture Considering different clusters, the message types are cat- egorized as follows. A message that is transmitted within a cluster is called internal, while a message that is transmitted across clusters is called external. The master nodes schedule the respective traffic on-line according to any desired scheduling policy (e.g., Fixed Priority Scheduling), on a cyclic basis. The basic cycle has a fixed duration of time and it is called Elementary Cycle (EC). In the hybrid architecture, each EC is partitioned among the traf- fic types, i.e., internal/external and synchronous/asynchronous traffic (Fig. 2). The external asynchronous window is further split into cluster sub-windows. The scheduler in the master node computes the activation instants of the synchronous messages and schedules them EC