Les plateformes de calcul à grande échelle ont beaucoup évoluées dernières années. La réduction des coûts des composants simplifie la construction de machines possédant des multicœurs et des accélérateurs comme les GPU.Ceci a permis une propagation des plateformes à grande échelle,dans lesquelles les machines peuvent être éloignées les unes des autres, pouvant même être situées sur différents continents. Le problème essentiel devient alors d'utiliser ces ressources efficacement.Dans ce travail nous nous intéressons d'abord à l'allocation efficace de tâches sur plateformes hétérogènes composées CPU et de GPU. Pour ce faire, nous proposons un outil nommé SWDUAL qui implémente l'algorithme de Smith-Waterman simultanément sur CPU et GPU, en choisissant quelles tâches il est plus intéressant de placer sur chaque type de ressource. Nos expériences montrent que SWDUAL donne de meilleurs résultats que les approches similaires de l'état de l'art.Nous analysons ensuite une nouvelle méthode d'ordonnancement enligne de tâches indépendantes de différentes tailles. Nous proposons une nouvelle technique qui optimise la métrique du stretch. Elle consiste à déplacer les jobs qui retardent trop de petites tâches sur des machines dédiées. Nos résultats expérimentaux montrent que notre méthode obtient de meilleurs résultats que la politique standard et qu'elle s'approche dans de nombreux cas des résultats d'une politique préemptive, qui peut être considérée comme une borne inférieure.Nous nous intéressons ensuite à l'impact de différentes contraintes sur la politique FCFS avec backfilling. La contrainte de contiguïté essaye de compacter les jobs et de réduire la fragmentation dans l'ordonnancement. La contrainte de localité basique place les jobs de telle sorte qu'ils utilisent le plus petit nombre de groupes de processeurs appelés textit. Nos résultats montrent que les bénéfices de telles contraintes sont suffisants pour compenser la réduction du nombre de jobs backfillés due à la réduction de la fragmentation.Nous proposons enfin une nouvelle contrainte nommée localité totale, dans laquelle l'ordonnanceur modélise la plateforme par un fat tree et se sert de cette information pour placer les jobs là où leur coût de communication est minimal.Notre campagne d'expériences montre que cette contrainte obtient de très bons résultats par rapport à un backfilling basique, et de meilleurs résultats que les contraintes précédentes. / The field of parallel supercomputing has been changing rapidly inrecent years. The reduction of costs of the parts necessary to buildmachines with multicore CPUs and accelerators such as GPUs are ofparticular interest to us. This scenario allowed for the expansion oflarge parallel systems, with machines far apart from each other,sometimes even located on different continents. Thus, the crucialproblem is how to use these resources efficiently.In this work, we first consider the efficient allocation of taskssuitable for CPUs and GPUs in heterogeneous platforms. To that end, weimplement a tool called SWDUAL, which executes the Smith-Watermanalgorithm simultaneously on CPUs and GPUs, choosing which tasks aremore suited to one or another. Experiments show that SWDUAL givesbetter results when compared to similar approaches available in theliterature.Second, we study a new online method for scheduling independent tasksof different sizes on processors. We propose a new technique thatoptimizes the stretch metric by detecting when a reasonable amount ofsmall jobs is waiting while a big job executes. Then, the big job isredirected to separate set of machines, dedicated to running big jobsthat have been redirected. We present experiment results that show thatour method outperforms the standard policy and in many cases approachesthe performance of the preemptive policy, which can be considered as alower bound.Next, we present our study on constraints applied to the Backfillingalgorithm in combination with the FCFS policy: Contiguity, which is aconstraint that tries to keep jobs close together and reducefragmentation during the schedule, and Basic Locality, that aims tokeep jobs as much as possible inside groups of processors calledclusters. Experiment results show that the benefits of using theseconstrains outweigh the possible decrease in the number of backfilledjobs due to reduced fragmentation.Finally, we present an additional constraint to the Backfillingalgorithm called Full Locality, where the scheduler models the topologyof the platform as a fat tree and uses this model to assign jobs toregions of the platform where communication costs between processors isreduced. The experiment campaign is executed and results show that FullLocality is superior to all the previously proposed constraints, andspecially Basic Backfilling.

<p>Switched Ethernet is an Ethernet LAN that uses switches to connect individual nodes. </p><p>This is popular because of its effective and convenient way of extending the bandwidth of </p><p>existing Ethernets. Switched Ethernet is being considered by the industry community because </p><p>of its open standardization, cost effectiveness, and the support for higher data rates up to </p><p>10Gbps. Even though many special-purposed solutions were proposed to support time </p><p>constrained communication over Switched Ethernet, still there were some doubts about the </p><p>real time handling capability of Switched Ethernet. To achieve reliable transmission </p><p>guarantees for real time traffic over Switched Ethernet, it is important to measure the </p><p>performance of Switched Ethernet networks for real time communication. In this thesis work </p><p>we have observed the average end-to-end packet delay for real time traffic over a Switched </p><p>Ethernet by simulation, which is very much essential for real time communication in </p><p>industrial applications, where the communication is time-deterministic. In our thesis we used </p><p>FCFS priority queuing in both the source nodes and switch. In this thesis we also discussed </p><p>about the feasibility analysis for fixed sized frames and some traffic handling methods. We </p><p>used 100mbp/s single full duplex Ethernet switch for our simulation. Finally simulation </p><p>analysis and simulation results are discussed. Our purpose of simulation of Switched Ethernet </p><p>networks is of good importance for the real time industrial applications.</p>

<p>In the past 10 years, the bandwidths and processing capabilities of the networks have increased dramatically. The number of real-time applications using these networks has also increased. The large number of real-time packets might, in a switched multi-hop network, lead to unpredictable traffic patterns. This is not a problem when the traffic intensity is low, but if the same network is used by a large number of users simultaneously, the overall performance of the network degrades. In fact, unpredictable delays in the delivery of the message can adversely affect the execution of the tasks dependent on these messages, even if we take into account the soft real-time performance.</p><p>In this paper, we investigate the effect of traffic shapers on soft real-time performance. We will consider a switched multi-hop network with FCFS queues. We will implement two versions of the network simulator. One version will be without traffic shaper and the other version will use a traffic shaper. By comparing the results (for average delay, deadline miss ratio etc.) from both the versions, we will try to conclude if it is really beneficial to use traffic shapers for soft real-time performance. Leaky bucket and token bucket algorithms are the most popular ones for traffic shaper implementation. We will consider leaky bucket algorithm for our analysis. We analyse different versions of the leaky bucket and present the trade-off’s involved.</p>

In the past 10 years, the bandwidths and processing capabilities of the networks have increased dramatically. The number of real-time applications using these networks has also increased. The large number of real-time packets might, in a switched multi-hop network, lead to unpredictable traffic patterns. This is not a problem when the traffic intensity is low, but if the same network is used by a large number of users simultaneously, the overall performance of the network degrades. In fact, unpredictable delays in the delivery of the message can adversely affect the execution of the tasks dependent on these messages, even if we take into account the soft real-time performance. In this paper, we investigate the effect of traffic shapers on soft real-time performance. We will consider a switched multi-hop network with FCFS queues. We will implement two versions of the network simulator. One version will be without traffic shaper and the other version will use a traffic shaper. By comparing the results (for average delay, deadline miss ratio etc.) from both the versions, we will try to conclude if it is really beneficial to use traffic shapers for soft real-time performance. Leaky bucket and token bucket algorithms are the most popular ones for traffic shaper implementation. We will consider leaky bucket algorithm for our analysis. We analyse different versions of the leaky bucket and present the trade-off’s involved.

Switched Ethernet is an Ethernet LAN that uses switches to connect individual nodes. This is popular because of its effective and convenient way of extending the bandwidth of existing Ethernets. Switched Ethernet is being considered by the industry community because of its open standardization, cost effectiveness, and the support for higher data rates up to 10Gbps. Even though many special-purposed solutions were proposed to support time constrained communication over Switched Ethernet, still there were some doubts about the real time handling capability of Switched Ethernet. To achieve reliable transmission guarantees for real time traffic over Switched Ethernet, it is important to measure the performance of Switched Ethernet networks for real time communication. In this thesis work we have observed the average end-to-end packet delay for real time traffic over a Switched Ethernet by simulation, which is very much essential for real time communication in industrial applications, where the communication is time-deterministic. In our thesis we used FCFS priority queuing in both the source nodes and switch. In this thesis we also discussed about the feasibility analysis for fixed sized frames and some traffic handling methods. We used 100mbp/s single full duplex Ethernet switch for our simulation. Finally simulation analysis and simulation results are discussed. Our purpose of simulation of Switched Ethernet networks is of good importance for the real time industrial applications.

<p>Diss. Stockholm :  Stockholm School of Economics, 2012. Introduction together with 6 papers</p>

Embedded applications have become more and more complex, increasing the demands on the communication network. For reasons such as safety and usability, there are real-time constraints that must be met. Also, to offer high performance, network protocols should offer efficient user services aimed at specific types of communication. At the same time, it is desirable to design and implement embedded networks with reduced cost and development time, which means using available hardware for standard networks. To that end, there is a trend towards using switched Ethernet for embedded systems because of its hight bit rate and low cost. Unfortunately, since switched Ethernet is not specifically designed for embedded systems, it has several limitations such as poor support for QoS because of FCFS queuing policy and high protocol overhead. This thesis contributes towards fulfilling these requirements by developing (i) real-time analytical frameworks for providing QoS guarantees in packet-switched networks and (II) packet-merging techniques to reduce the protocol overhead. We have developed two real-time analytical frameworks for networks with FCFS queuing in the switches, one for FCFS queuing in the source nodes and one for EDF queuing in the source nodes. The correctness and tightness of the real-time analytical frameworks for different network components in a singel-switch neetwork are given by strict theoretical proofs, and the performance of our end-to-end analyses is evaluated by simulations. In conjunction with this, we have compared our results to Network Calculus (NC), a commonly used analytical scheme for FCFS queuing. Our comparison study shows that our anlysis is more accurate than NC for singel-switch networks. To reduce the protocol overhead, we have proposed two active switched Ethernet approaches, one for real-time many-to-many communication and the other for the real-time short message traffic that is often present in embedded applications. A significant improvement in performance achieved by using our proposed active networks is demonstrated. Although our approaches are exemplified using switched Ethernet, the general approaches are not limited to switched Ethernet networks but can easily be moified to other similar packet-switched networks.

The growing interest towards wireless communication services over the recent years has increased the demand for radio spectrum. Inefficient spectrum management together with the scarcity of the radio spectrum is a limiting factor for the development of modern wireless networks. As a solution, the idea of cognitive radio networks (CRNs) is introduced to use licensed spectrum for the benefit of the unlicensed secondary users. However, the preemptive priority of the licensed users results in random resource availabilities at the secondary networks, which makes the quality-of-service (QoS) support challenging. With the increasing demand for elastic/interactive data services (internet based services) and wireless multimedia services, QoS support becomes essential for CRNs. This research investigates the voice and elastic/interactive data service support over CRNs, in terms of their delay requirements. The packet level requirements of the voice service and session level delay requirements of the elastic/interactive data services are studied. In particular, constant-rate and on-off voice traffic capacities are analyzed over CRNs with centralized and distributed network coordination. Some generic channel access schemes are considered as the coordination mechanism, and call admission control algorithms are developed for non-fully-connected CRNs. Advantage of supporting voice traffic flows with different delay requirements in the same network is also discussed. The mean response time of the elastic data traffic over a centralized CRN is studied, considering the shortest processor time with and without preemption and shortest remaining processor time service disciplines, in comparison with the processor sharing service discipline. Effects of the traffic load at the base station and file length (service time requirement) distribution on the mean response time are discussed. Finally, the relationship between the mean response times of interactive and elastic data traffic is studied.

The growing interest towards wireless communication services over the recent years has increased the demand for radio spectrum. Inefficient spectrum management together with the scarcity of the radio spectrum is a limiting factor for the development of modern wireless networks. As a solution, the idea of cognitive radio networks (CRNs) is introduced to use licensed spectrum for the benefit of the unlicensed secondary users. However, the preemptive priority of the licensed users results in random resource availabilities at the secondary networks, which makes the quality-of-service (QoS) support challenging. With the increasing demand for elastic/interactive data services (internet based services) and wireless multimedia services, QoS support becomes essential for CRNs. This research investigates the voice and elastic/interactive data service support over CRNs, in terms of their delay requirements. The packet level requirements of the voice service and session level delay requirements of the elastic/interactive data services are studied. In particular, constant-rate and on-off voice traffic capacities are analyzed over CRNs with centralized and distributed network coordination. Some generic channel access schemes are considered as the coordination mechanism, and call admission control algorithms are developed for non-fully-connected CRNs. Advantage of supporting voice traffic flows with different delay requirements in the same network is also discussed. The mean response time of the elastic data traffic over a centralized CRN is studied, considering the shortest processor time with and without preemption and shortest remaining processor time service disciplines, in comparison with the processor sharing service discipline. Effects of the traffic load at the base station and file length (service time requirement) distribution on the mean response time are discussed. Finally, the relationship between the mean response times of interactive and elastic data traffic is studied.

Virtualisation is one of the underlying technologies that led to the success of cloud computing platforms (CCPs). The technology, along with other features such as multitenancy allows delivering of computing resources in the form of service through efficient sharing of physical resources. As these resources are provided through virtualisation, a robust agreement is outlined for both the quantity and quality-of-service (QoS) in a service level agreement (SLA) documents. QoS is one of the essential components of SLA, where performance is one of its primary aspects. As the technology is progressively maturing and receiving massive acceptance, researchers from industry and academia continue to carry out novel theoretical and practical studies of various essential aspects of CCPs with significant levels of success. This thesis starts with the assessment of the current level of knowledge in the literature of cloud computing in general and CCPs in particular. In this context, a substantive literature review was carried out focusing on performance modelling, testing, analysis and evaluation of Infrastructure as a Service (IaaS), methodologies. To this end, a systematic mapping study (SMSs) of the literature was conducted. SMS guided the choice and direction of this research. The SMS was followed by the development of a novel open queueing network model (QNM) at equilibrium for the performance modelling and analysis of an OpenStack IaaS CCP. Moreover, it was assumed that an external arrival pattern is Poisson while the queueing stations provided exponentially distributed service times. Based on Jackson’s theorem, the model was exactly decomposed into individual M/M/c (c ≥ 1) stations. Each of these queueing stations was analysed in isolation, and closed-form expressions for key performance metrics, such as mean response time, throughput, server (resource) utilisation as well as bottleneck device were determined. Moreover, the research was extended with a proposed open QNM with a bursty external arrival pattern represented by a Compound Poisson Process (CPP) with geometrically distributed batches, or equivalently, variable Generalised Exponential (GE) interarrival and service times. Each queueing station had c (c ≥ 1) GE-type servers. Based on a generic maximum entropy (ME) product form approximation, the proposed open GE-type QNM was decomposed into individual GE/GE/c queueing stations with GE-type interarrival and service times. The evaluation of the performance metrics and bottleneck analysis of the QNM were determined, which provided vital insights for the capacity planning of existing CCP architectures as well as the design and development of new ones. The results also revealed, due to a significant impact on the burstiness of interarrival and service time processes, resulted in worst-case performance bounds scenarios, as appropriate. Finally, an investigation was carried out into modelling and analysis of performance and security trade-offs for a CCP architecture, based on a proposed generalised stochastic Petri net (GSPN) model with security-detection control model (SDCM). In this context, ‘optimal’ combined performance and security metrics were defined with both M-type or GE-type arrival and service times and the impact of security incidents on performance was assessed. Typical numerical experiments on the GSPN model were conducted and implemented using the Möbius package, and an ‘optimal’ trade-offs were determined between performance and security, which are crucial in the SLA of the cloud computing services. / Petroleum technology development fund (PTDF) of the government of Nigeria Usmanu Danfodiyo University, Sokoto