On the Scalability-Performance Tradeoffs in MPLS and IP Routing

MPLS (Multi-Protocol Label Switching) has recently emerged to facilitate the engineering of network traffic. This can be achieved by directing packet flows over paths that satisfy multiple requirements. MPLS has been regarded as an enhancement to traditional IP routing, which has the following problems: (1) all packets with the same IP destination address have to follow the same path through the network; and (2) paths have often been computed based on static and single link metrics. These problems may cause traffic concentration, and thus degradation in quality of service. In this paper, we investigate by simulations a range of routing solutions and examine the tradeoff between scalability and performance. At one extreme, IP packet routing using dynamic link metrics provides a stateless solution but may lead to routing oscillations. At the other extreme, we consider a recently proposed Profile-based Routing (PBR), which uses knowledge of potential ingress-egress pairs as well as the traffic profile among them. Minimum Interference Routing (MIRA) is another recently proposed MPLS-based scheme, which only exploits knowledge of potential ingress-egress pairs but not their traffic profile. MIRA and the more conventional widest-shortest path (WSP) routing represent alternative MPLS-based approaches on the spectrum of routing solutions. We compare these solutions in terms of utility, bandwidth acceptance ratio as well as their scalability (routing state and computational overhead) and load balancing capability. While the simplest of the per-flow algorithms we consider, the performance of WSP is close to dynamic per-packet routing, without the potential instabilities of dynamic routing.

Cluster-Based Optimizations for Distributed Hash Tables

We consider the problem of performing topological optimizations of distributed hash tables. Such hash tables include Chord and Tapestry and are a popular building block for distributed applications. Optimizing topologies over one dimensional hash spaces is particularly difficult as the higher dimensionality of the underlying network makes close fits unlikely. Instead, current schemes are limited to heuristically performing local optimizations finding the best of small random set of peers. We propose a new class of topology optimizations based on the existence of clusters of close overlay members within the underlying network. By constructing additional overlays for each cluster, a significant portion of the search procedure can be performed within the local cluster with a corresponding reduction in the search time. Finally, we discuss the effects of these additional overlays on spatial locality and other load balancing scheme.

An Infrastructure for the Dynamic Distribution of Web Applications and Services

This paper presents the design and implementation of an infrastructure that enables any Web application, regardless of its current state, to be stopped and uninstalled from a particular server, transferred to a new server, then installed, loaded, and resumed, with all these events occurring "on the fly" and totally transparent to clients. Such functionalities allow entire applications to fluidly move from server to server, reducing the overhead required to administer the system, and increasing its performance in a number of ways: (1) Dynamic replication of new instances of applications to several servers to raise throughput for scalability purposes, (2) Moving applications to servers to achieve load balancing or other resource management goals, (3) Caching entire applications on servers located closer to clients.

Multiphase mesh partitioning

We consider the load-balancing problems which arise from parallel scientific codes containing multiple computational phases, or loops over subsets of the data, which are separated by global synchronisation points. We motivate, derive and describe the implementation of an approach which we refer to as the multiphase mesh partitioning strategy to address such issues. The technique is tested on several examples of meshes, both real and artificial, containing multiple computational phases and it is demonstrated that our method can achieve high quality partitions where a standard mesh partitioning approach fails.

Multiphase mesh partitioning for parallel computational mechanics codes

We consider the load-balancing problems which arise from parallel scientific codes containing multiple computational phases, or loops over subsets of the data, which are separated by global synchronisation points. We motivate, derive and describe the implementation of an approach which we refer to as the multiphase mesh partitioning strategy to address such issues. The technique is tested on example meshes containing multiple computational phases and it is demonstrated that our method can achieve high quality partitions where a standard mesh partitioning approach fails.

The development of parallel implementation for a CFD based fire field model utilising conventional office based PCs

Parallel processing techniques have been used in the past to provide high performance computing resources for activities such as fire-field modelling. This has traditionally been achieved using specialized hardware and software, the expense of which would be difficult to justify for many fire engineering practices. In this article we demonstrate how typical office-based PCs attached to a Local Area Network has the potential to offer the benefits of parallel processing with minimal costs associated with the purchase of additional hardware or software. It was found that good speedups could be achieved on homogeneous networks of PCs, for example a problem composed of ~100,000 cells would run 9.3 times faster on a network of 12 800MHz PCs than on a single 800MHz PC. It was also found that a network of eight 3.2GHz Pentium 4 PCs would run 7.04 times faster than a single 3.2GHz Pentium computer. A dynamic load balancing scheme was also devised to allow the effective use of the software on heterogeneous PC networks. This scheme also ensured that the impact between the parallel processing task and other computer users on the network was minimized.

Parallel algorithms of the Purcell method for direct solution of linear systems

In this paper, we first demonstrate that the classical Purcell's vector method when combined with row pivoting yields a consistently small growth factor in comparison to the well-known Gauss elimination method, the Gauss–Jordan method and the Gauss–Huard method with partial pivoting. We then present six parallel algorithms of the Purcell method that may be used for direct solution of linear systems. The algorithms differ in ways of pivoting and load balancing. We recommend algorithms V and VI for their reliability and algorithms III and IV for good load balance if local pivoting is acceptable. Some numerical results are presented.

WEBCOM-G: A candidate middleware for Grid-Ireland

WebCom-G is a fledgling Grid Operating System, designed to provide independent service access through interoperability with existing middlewares. It offers an expressive programming model that automatically handles task synchronisation – load balancing, fault tolerance, and task allocation are handled at the WebCom-G system level – without burdening the application writer. These characteristics, together with the ability of its computing model to mix evaluation strategies to match the characteristics of the geographically dispersed facilities and the overall problem- solving environment, make WebCom-G a promising grid middleware candidate.

Ultra-fast processing of gigapixel Tissue MicroArray images using high performance computing.

BACKGROUND:
tissue MicroArrays (TMAs) are a valuable platform for tissue based translational research and the discovery of tissue biomarkers. The digitised TMA slides or TMA Virtual Slides, are ultra-large digital images, and can contain several hundred samples. The processing of such slides is time-consuming, bottlenecking a potentially high throughput platform.
METHODS:
a High Performance Computing (HPC) platform for the rapid analysis of TMA virtual slides is presented in this study. Using an HP high performance cluster and a centralised dynamic load balancing approach, the simultaneous analysis of multiple tissue-cores were established. This was evaluated on Non-Small Cell Lung Cancer TMAs for complex analysis of tissue pattern and immunohistochemical positivity.
RESULTS:
the automated processing of a single TMA virtual slide containing 230 patient samples can be significantly speeded up by a factor of circa 22, bringing the analysis time to one minute. Over 90 TMAs could also be analysed simultaneously, speeding up multiplex biomarker experiments enormously.
CONCLUSIONS:
the methodologies developed in this paper provide for the first time a genuine high throughput analysis platform for TMA biomarker discovery that will significantly enhance the reliability and speed for biomarker research. This will have widespread implications in translational tissue based research.

Compiler and Runtime Support for Hybrid Static/Dynamic Scheduling

One of the outstanding issues in parallel computing is the selection of task granularity. This work proposes a solution to the task granularity problem by lowering the overhead of the task scheduler and as such supporting very fine-grain tasks. Using a combination of static (compile-time) scheduling and dynamic (run-time) scheduling, we aim to make scheduling decisions as fast as with static scheduling while retaining the dynamic load- balancing properties of fully dynamic scheduling. We present an example application and discuss the requirements on the compiler and runtime system to realize hybrid static/dynamic scheduling.

Network virtualisation from an operator perspective

Network virtualisation is seen as a promising approach to overcome the so-called “Internet impasse” and bring innovation back into the Internet, by allowing easier migration towards novel networking approaches as well as the coexistence of complementary network architectures on a shared infrastructure in a commercial context. Recently, the interest from the operators and mainstream industry in network virtualisation has grown quite significantly, as the potential benefits of virtualisation became clearer, both from an economical and an operational point of view. In the beginning, the concept has been mainly a research topic and has been materialized in small-scale testbeds and research network environments. This PhD Thesis aims to provide the network operator with a set of mechanisms and algorithms capable of managing and controlling virtual networks. To this end, we propose a framework that aims to allocate, monitor and control virtual resources in a centralized and efficient manner. In order to analyse the performance of the framework, we performed the implementation and evaluation on a small-scale testbed. To enable the operator to make an efficient allocation, in real-time, and on-demand, of virtual networks onto the substrate network, it is proposed a heuristic algorithm to perform the virtual network mapping. For the network operator to obtain the highest profit of the physical network, it is also proposed a mathematical formulation that aims to maximize the number of allocated virtual networks onto the physical network. Since the power consumption of the physical network is very significant in the operating costs, it is important to make the allocation of virtual networks in fewer physical resources and onto physical resources already active. To address this challenge, we propose a mathematical formulation that aims to minimize the energy consumption of the physical network without affecting the efficiency of the allocation of virtual networks. To minimize fragmentation of the physical network while increasing the revenue of the operator, it is extended the initial formulation to contemplate the re-optimization of previously mapped virtual networks, so that the operator has a better use of its physical infrastructure. It is also necessary to address the migration of virtual networks, either for reasons of load balancing or for reasons of imminent failure of physical resources, without affecting the proper functioning of the virtual network. To this end, we propose a method based on cloning techniques to perform the migration of virtual networks across the physical infrastructure, transparently, and without affecting the virtual network. In order to assess the resilience of virtual networks to physical network failures, while obtaining the optimal solution for the migration of virtual networks in case of imminent failure of physical resources, the mathematical formulation is extended to minimize the number of nodes migrated and the relocation of virtual links. In comparison with our optimization proposals, we found out that existing heuristics for mapping virtual networks have a poor performance. We also found that it is possible to minimize the energy consumption without penalizing the efficient allocation. By applying the re-optimization on the virtual networks, it has been shown that it is possible to obtain more free resources as well as having the physical resources better balanced. Finally, it was shown that virtual networks are quite resilient to failures on the physical network.

Reliable and energy efficient routing for ad hoc networks

In Mobile Ad hoc NETworks (MANETs), where cooperative behaviour is mandatory, there is a high probability for some nodes to become overloaded with packet forwarding operations in order to support neighbor data exchange. This altruistic behaviour leads to an unbalanced load in the network in terms of traffic and energy consumption. In such scenarios, mobile nodes can benefit from the use of energy efficient and traffic fitting routing protocol that better suits the limited battery capacity and throughput limitation of the network. This PhD work focuses on proposing energy efficient and load balanced routing protocols for ad hoc networks. Where most of the existing routing protocols simply consider the path length metric when choosing the best route between a source and a destination node, in our proposed mechanism, nodes are able to find several routes for each pair of source and destination nodes and select the best route according to energy and traffic parameters, effectively extending the lifespan of the network. Our results show that by applying this novel mechanism, current flat ad hoc routing protocols can achieve higher energy efficiency and load balancing. Also, due to the broadcast nature of the wireless channels in ad hoc networks, other technique such as Network Coding (NC) looks promising for energy efficiency. NC can reduce the number of transmissions, number of re-transmissions, and increase the data transfer rate that directly translates to energy efficiency. However, due to the need to access foreign nodes for coding and forwarding packets, NC needs a mitigation technique against unauthorized accesses and packet corruption. Therefore, we proposed different mechanisms for handling these security attacks by, in particular by serially concatenating codes to support reliability in ad hoc network. As a solution to this problem, we explored a new security framework that proposes an additional degree of protection against eavesdropping attackers based on using concatenated encoding. Therefore, malicious intermediate nodes will find it computationally intractable to decode the transitive packets. We also adopted another code that uses Luby Transform (LT) as a pre-coding code for NC. Primarily being designed for security applications, this code enables the sink nodes to recover corrupted packets even in the presence of byzantine attacks.

Dynamic global scheduling of parallel real-time tasks

High-level parallel languages offer a simple way for application programmers to specify parallelism in a form that easily scales with problem size, leaving the scheduling of the tasks onto processors to be performed at runtime. Therefore, if the underlying system cannot efficiently execute those applications on the available cores, the benefits will be lost. In this paper, we consider how to schedule highly heterogenous parallel applications that require real-time performance guarantees on multicore processors. The paper proposes a novel scheduling approach that combines the global Earliest Deadline First (EDF) scheduler with a priority-aware work-stealing load balancing scheme, which enables parallel realtime tasks to be executed on more than one processor at a given time instant. Experimental results demonstrate the better scalability and lower scheduling overhead of the proposed approach comparatively to an existing real-time deadline-oriented scheduling class for the Linux kernel.

Real-Time Application Mapping for Many-Cores Using a Limited Migrative Model

Many-core platforms are an emerging technology in the real-time embedded domain. These devices offer various options for power savings, cost reductions and contribute to the overall system flexibility, however, issues such as unpredictability, scalability and analysis pessimism are serious challenges to their integration into the aforementioned area. The focus of this work is on many-core platforms using a limited migrative model (LMM). LMM is an approach based on the fundamental concepts of the multi-kernel paradigm, which is a promising step towards scalable and predictable many-cores. In this work, we formulate the problem of real-time application mapping on a many-core platform using LMM, and propose a three-stage method to solve it. An extended version of the existing analysis is used to assure that derived mappings (i) guarantee the fulfilment of timing constraints posed on worst-case communication delays of individual applications, and (ii) provide an environment to perform load balancing for e.g. energy/thermal management, fault tolerance and/or performance reasons.

Gestion adaptative des ressources dans les réseaux maillés sans fil à multiples-radios multiples-canaux

Depuis quelques années, la recherche dans le domaine des réseaux maillés sans fil ("Wireless Mesh Network (WMN)" en anglais) suscite un grand intérêt auprès de la communauté des chercheurs en télécommunications. Ceci est dû aux nombreux avantages que la technologie WMN offre, telles que l'installation facile et peu coûteuse, la connectivité fiable et l'interopérabilité flexible avec d'autres réseaux existants (réseaux Wi-Fi, réseaux WiMax, réseaux cellulaires, réseaux de capteurs, etc.). Cependant, plusieurs problèmes restent encore à résoudre comme le passage à l'échelle, la sécurité, la qualité de service (QdS), la gestion des ressources, etc. Ces problèmes persistent pour les WMNs, d'autant plus que le nombre des utilisateurs va en se multipliant. Il faut donc penser à améliorer les protocoles existants ou à en concevoir de nouveaux. L'objectif de notre recherche est de résoudre certaines des limitations rencontrées à l'heure actuelle dans les WMNs et d'améliorer la QdS des applications multimédia temps-réel (par exemple, la voix). Le travail de recherche de cette thèse sera divisé essentiellement en trois principaux volets: le contrôle d‟admission du trafic, la différentiation du trafic et la réaffectation adaptative des canaux lors de la présence du trafic en relève ("handoff" en anglais). Dans le premier volet, nous proposons un mécanisme distribué de contrôle d'admission se basant sur le concept des cliques (une clique correspond à un sous-ensemble de liens logiques qui interfèrent les uns avec les autres) dans un réseau à multiples-sauts, multiples-radios et multiples-canaux, appelé RCAC. Nous proposons en particulier un modèle analytique qui calcule le ratio approprié d'admission du trafic et qui garantit une probabilité de perte de paquets dans le réseau n'excédant pas un seuil prédéfini. Le mécanisme RCAC permet d‟assurer la QdS requise pour les flux entrants, sans dégrader la QdS des flux existants. Il permet aussi d‟assurer la QdS en termes de longueur du délai de bout en bout pour les divers flux. Le deuxième volet traite de la différentiation de services dans le protocole IEEE 802.11s afin de permettre une meilleure QdS, notamment pour les applications avec des contraintes temporelles (par exemple, voix, visioconférence). À cet égard, nous proposons un mécanisme d'ajustement de tranches de temps ("time-slots"), selon la classe de service, ED-MDA (Enhanced Differentiated-Mesh Deterministic Access), combiné à un algorithme efficace de contrôle d'admission EAC (Efficient Admission Control), afin de permettre une utilisation élevée et efficace des ressources. Le mécanisme EAC prend en compte le trafic en relève et lui attribue une priorité supérieure par rapport au nouveau trafic pour minimiser les interruptions de communications en cours. Dans le troisième volet, nous nous intéressons à minimiser le surcoût et le délai de re-routage des utilisateurs mobiles et/ou des applications multimédia en réaffectant les canaux dans les WMNs à Multiples-Radios (MR-WMNs). En premier lieu, nous proposons un modèle d'optimisation qui maximise le débit, améliore l'équité entre utilisateurs et minimise le surcoût dû à la relève des appels. Ce modèle a été résolu par le logiciel CPLEX pour un nombre limité de noeuds. En second lieu, nous élaborons des heuristiques/méta-heuristiques centralisées pour permettre de résoudre ce modèle pour des réseaux de taille réelle. Finalement, nous proposons un algorithme pour réaffecter en temps-réel et de façon prudente les canaux aux interfaces. Cet algorithme a pour objectif de minimiser le surcoût et le délai du re-routage spécialement du trafic dynamique généré par les appels en relève. Ensuite, ce mécanisme est amélioré en prenant en compte l‟équilibrage de la charge entre cliques.