Eric Bouillet

Efficient resource provisioning in compute clouds via VM multiplexing

Resource provisioning in compute clouds often require an estimate of the capacity needs of Virtual Machines (VMs). The estimated VM size is the basis for allocating resources commensurate with workload demand. In contrast to the traditional practice of estimating the VM sizes individually, we propose a joint-VM sizing approach in which multiple VMs are consolidated and provisioned, based on an estimate of their aggregate capacity needs. This new approach exploits statistical multiplexing among the workload patterns of multiple VMs, i.e., the peaks and valleys in one workload pattern do not necessarily coincide with the others. Thus, the unused resources of a low utilized VM can be directed to the other co-located VMs with high utilization. Compared to individual VM based provisioning, joint-VM sizing and provisioning may lead to much higher resource utilization. This paper presents three design modules to enable the concept in practice. Specifically, a performance constraint describing the capacity need of a VM for achieving a certain level of application performance; an algorithm for estimating the size of jointly provisioning VMs; a VM selection method that seeks to find good VM combinations for being provisioned together. We showcase that the proposed three modules can be seamlessly plugged into existing applications such as resource provisioning, and providing resource guarantees for VMs. The proposed algorithms and applications are evaluated by monitoring data collected from about 16 thousand VMs in commercial data centers. These evaluations reveal more than 45% improvements in terms of the overall resource utilization.

IBM infosphere streams for scalable, real-time, intelligent transportation services

With the widespread adoption of location tracking technologies like GPS, the domain of intelligent transportation services has seen growing interest in the last few years. Services in this domain make use of real-time location-based data from a variety of sources, combine this data with static location-based data such as maps and points of interest databases, and provide useful information to end-users. Some of the major challenges in this domain include i) scalability, in terms of processing large volumes of real-time and static data; ii) extensibility, in terms of being able to add new kinds of analyses on the data rapidly, and iii) user interaction, in terms of being able to support different kinds of one-time and continuous queries from the end-user. In this paper, we demonstrate the use of IBM InfoSphere Streams, a scalable stream processing platform, for tackling these challenges. We describe a prototype system that generates dynamic, multi-faceted views of transportation information for the city of Stockholm, using real vehicle GPS and road-network data. The system also continuously derives current traffic statistics, and provides useful value-added information such as shortest-time routes from real-time observed and inferred traffic conditions. Our performance experiments illustrate the scalability of the system. For instance, our system can process over 120000 incoming GPS points per second, combine it with a map containing over 600,000 links, continuously generate different kinds of traffic statistics and answer user queries.

Lightpath re-optimization in mesh optical networks

Intelligent mesh optical networks deployed today offer unparalleled capacity, flexibility, availability, and, inevitably, new challenges to master all these qualities in the most efficient and practical manner. More specifically, demands are routed according to the state of the network available at the moment. As the network and the traffic evolve, the lightpaths of the existing demands becomes sub-optimal. In this paper we study two algorithms to re-optimize lightpaths in resilient mesh optical networks. One is a complete re-optimization algorithm that re-routes both primary and backup paths, and the second is a partial re-optimization algorithm that re-routes the backup paths only. We show that on average, these algorithms allow bandwidth savings of 3% to 5% of the total capacity in scenarios where the backup path only is re-routed, and substantially larger bandwidth savings when both the working and backup paths are re-routed. We also prove that trying all possible demand permutations with an online algorithm does not guarantee optimality, and in certain cases does not achieve it, while for the same scenario optimality is achieved through re-optimization. This observation motivates the needs for a re-optimization approach that does not just simply look at different sequences, and we propose and experiment with such an approach. Re-optimization has actually been performed in a nationwide live optical mesh network and the resulting savings are reported in this paper, validating reality and the usefulness of re-optimization in real networks.

Stochastic approaches to compute shared mesh restored lightpaths in optical network architectures

We assess the benefits of using statistical techniques to ascertain the shareability of protection channels when computing shared mesh restored lightpaths. Current deterministic approaches require a detailed level of information proportional to the number of active lightpaths, and do not scale well as traffic demands and network grow. With the proposed approach, we show that less information, independent of the amount of traffic demand, is sufficient to determine the shareability of protection channels with remarkable accuracy. Experiments also demonstrate that our approach yields faster computation times with no significant penalty in terms of capacity usage.

Path Routing in Mesh Optical Networks

List of Figures. List of Tables. Foreword. Preface. 1 Optical Networking. 1.1 Evolution of Optical Network Architectures. 1.1.1 Transparent Networks. 1.1.2 Opaque Networks. 1.1.3 Translucent Networks. 1.2 Layered Network Architecture. 1.2.1 Optical Layer. 1.2.2 Logical Layer. 1.2.3 Service/Application Layer. 1.3 Multi-Tier Optical Layer. 1.3.1 One-Tier Network Architecture. 1.3.2 Two-Tier Network Architecture. 1.3.3 Network Scalability. 1.4 The Current State of Optical Networks. 1.5 Organization of the Book. 2 Recovery in Optical Networks. 2.1 Introduction. 2.2 Failure Recovery. 2.3 Fault Recovery Classifications. 2.4 Protection of Point-to-Point Systems. 2.4.1 (1 + 1) Protection. 2.4.2 (1 : 1) Protection. 2.4.3 (M :N) Protection. 2.5 Ring-Based Protection. 2.5.1 Failure Recovery in SONET Networks with Ring Topologies. 2.5.2 Ring-Based Failure Recovery in Optical Networks with Mesh Topologies. 2.6 Path-Based Protection. 2.6.1 Dedicated Backup Path Protection (DBPP) in Mesh Networks. 2.6.2 Shared Back Path Protection (SBPP) in Mesh Networks. 2.7 Link/Span-Based Protection. 2.8 Segment-Based Protection. 2.9 Island-Based Protection. 2.10 Mesh Network Restoration. 2.10.1 Centralized Restoration Techniques. 2.10.2 Distributed Restoration Techniques. 2.11 Multi-Layer Recovery. 2.12 Recovery Triggers and Signaling Mechanisms. 2.13 Conclusion. 3 Mesh Routing and Recovery Framework. 3.1 Introduction. 3.2 Mesh Protection and Recovery Techniques. 3.2.1 Link-Based Protection. 3.2.2 Path-Based Protection. 3.2.3 Segment-Based Protection. 3.3 Concept of Shared Risk Groups. 3.3.1 Shared Link Risk Groups. 3.3.2 Shared Node Risk Groups. 3.3.3 Shared Equipment Risk Groups. 3.4 Centralized vs Distributed Routing. 3.4.1 Centralized Routing. 3.4.2 Distributed Routing. 3.4.3 Centralized vs Distributed Routing Performance Results. 3.5 Conclusion. 4 Path Routing and Protection. 4.1 Introduction. 4.2 Routing in Path-Protected Mesh Networks. 4.3 Protection in Path-Protected Mesh Networks. 4.3.1 Dedicated Backup Path-Protected Lightpaths. 4.3.2 Shared Backup Path-Protected Lightpaths. 4.3.3 Preemptible Lightpaths. 4.3.4 Diverse Unprotected Lightpaths with Dual-Homing. 4.3.5 Multiple Simultaneous Backup Path-Protected Lightpaths. 4.3.6 Relaxing the Protection Guarantees. 4.3.7 Impact of Multi-Port Card Diversity Constraints. 4.4 Experiments and Capacity Performance Results. 4.4.1 Performance Results for Path-Based Protection Techniques. 4.4.2 Experiments with Multi-Port Card Diversity. 4.5 Recovery Time Analysis. 4.6 Recovery Time and Capacity Trade-Offs. 4.7 Conclusion. 5 Path Routing - Part 1: Complexity. 5.1 Introduction. 5.2 Network Topology Abstraction. 5.2.1 Service Definition. 5.2.2 Operational Models: Online vs Offline Routing. 5.3 Shortest-Path Routing. 5.3.1 Dijkstras Algorithm. 5.3.2 Dijkstras Algorithm Generalization to K-Shortest Paths. 5.3.3 Shortest-Path Routing with Constraints. 5.4 Diverse-Path Routing. 5.4.1 SRG Types. 5.4.2 Diverse-Path Routing with Default SRGs. 5.4.3 Diverse-Path Routing with Fork SRGs. 5.4.4 Diverse-Path Routing with General SRGs. 5.5 Shared Backup Path Protection Routing. 5.5.1 Protection Guarantees and Rules of Sharing. 5.5.2 Complexity of Shared Backup Path Protection Routing. 5.6 Routing ILP. 5.6.1 ILP Description. 5.6.2 Implementation Experience. 5.7 Conclusion. 5.8 Appendix. 5.8.1 Complexity of Diverse-Path Routing with General SRGs. 5.8.2 Complexity of SBPP Routing. 6 Path Routing - Part 2: Heuristics. 6.1 Introduction. 6.1.1 Operational Models: Centralized vs Distributed Routing. 6.1.2 Topology Modeling Example. 6.2 Motivating Problems. 6.2.1 Heuristic Techniques. 6.3 K-Shortest Path Routing. 6.3.1 Yens K-Shortest Path Algorithm. 6.3.2 Constrained Shortest-Path Routing. 6.4 Diverse-Path Routing. 6.4.1 Best-Effort Path Diversity. 6.5 Shared Backup Path Protection Routing. 6.5.1 Sharing-Independent Routing Heuristic. 6.5.2 Sharing-Dependent Routing Heuristic. 6.6 Routing Preemptible Services. 6.7 General Constrained Routing Framework. 6.7.1 Implementation Experience. 6.8 Conclusion. 7 Enhanced Routing Model for SBPP Services. 7.1 Introduction. 7.2 Routing Metric. 7.3 Routing Algorithm. 7.4 Experiments. 7.4.1 Effect of . 7.4.2 Effect of alpha. 7.5 Conclusion. 8 Controlling Sharing for SBPP Services. 8.1 Introduction. 8.2 Express Links. 8.2.1 Routing with Express Links. 8.2.2 Analysis and Results. 8.2.3 Express Links-Conclusion. 8.3 Limiting Sharing. 8.3.1 Example. 8.3.2 Solution Alternatives. 8.3.3 Analysis of Capping. 8.3.4 Analysis of Load-Balancing. 8.3.5 Limiting Sharing-Conclusion. 8.4 Analysis of Active Reprovisioning. 8.4.1 Evaluation of Active Reprovisioning. 8.4.2 Active Reprovisioning-Conclusion. 8.5 Conclusion. 9 Path Computation with Partial Information. 9.1 Introduction. 9.2 Complexity of the Deterministic Approach. 9.2.1 Complexity of the Failure Dependent Strategy. 9.2.2 Complexity of the Failure Independent Strategy. 9.3 Probabilistic Approach. 9.3.1 A Problem of Combinations. 9.3.2 Analogy with SRG Arrangement into a Set of Backup Channels. 9.4 Probabilistic Routing Algorithm with Partial Information. 9.5 Locally Optimized Channel Selection. 9.5.1 Shared Mesh Protection Provisioning Using Vertex Coloring. 9.5.2 Implementation and Applications. 9.6 Required Extensions to Routing Protocols. 9.7 Experiments and Performance Results. 9.7.1 Accuracy and Distributions of Probability Functions. 9.7.2 Comparison of Deterministic vs ProbabilisticWeight Functions on Real Networks. 9.7.3 Benefits of Locally Optimized Lightpath Provisioning. 9.7.4 Summary. 9.8 Conclusion. 10 Path Reoptimization. 10.1 Introduction. 10.2 Routing Algorithm. 10.2.1 Cost model. 10.2.2 Online Routing Algorithm. 10.3 Reoptimization Algorithm. 10.4 The Complexity of Reoptimization. 10.4.1 No Prior Placement of Protection Channels or Primary Paths. 10.4.2 Prior Placement of Protection Channels or Primary Paths. 10.5 Experiments. 10.5.1 Calibration. 10.5.2 Real Networks. 10.5.3 Static Network Infrastructure. 10.5.4 Growing Network Infrastructure. 10.5.5 Network Dynamics. 10.6 Conclusion. 11 Dimensioning of Path-Protected Mesh Networks. 11.1 Introduction. 11.2 Network and Traffic Modeling. 11.3 Mesh Network Characteristics. 11.3.1 Path Length Analysis. 11.3.2 Protection-to-Working Capacity Ratio Analysis. 11.3.3 Sharing Analysis. 11.4 Asymptotic Behavior of the Protection-to-Working Capacity Ratio. 11.4.1 Examples. 11.4.2 General Results. 11.5 Dimensioning Mesh Optical Networks. 11.5.1 Node Model and Traffic Conservation Equations. 11.5.2 Dimensioning Examples and Results. 11.6 The Network Global Expectation Model. 11.7 Accuracy of Analytical Estimates. 11.8 Recovery Time Performance. 11.9 Conclusion. 12 Service Availability in Path-Protected Mesh Networks. 12.1 Introduction. 12.2 Network Service Availability. 12.2.1 Motivation. 12.2.2 Focus on Dual-Failure Scenarios. 12.2.3 Reliability and Availability. 12.3 Service Availability in Path-Protected Mesh Networks. 12.3.1 Dual-Failure Recoverability. 12.3.2 A Markov Model Approach to Service Availability. 12.3.3 Modeling Sharing of Backup Channels. 12.3.4 Impact of Channel Protection. 12.3.5 Impact of Reprovisioning. 12.4 Availability in Single and Multiple Domains. 12.4.1 Network Recovery Architecture-Single Domain. 12.4.2 Network Recovery Architecture-Multiple Domains. 12.4.3 Results and Discussion. 12.4.4 A Simple Model. 12.5 Availability in Ring and Path-Protected Networks. 12.5.1 Ring Availability Analysis. 12.5.2 Results and Discussion. 12.5.3 The Simple Model Again. 12.6 Conclusion. Bibliography. Index.

A Folksonomy-Based Model of Web Services for Discovery and Automatic Composition

In this paper, we propose a novel way of modeling Web services using folksonomies. The key advantage of our model is that it allows a large number of users to participate, easily, in annotating services with tags. This is in contrast to more expressive, logic based models of services, such as semantic Web service models, which require significant expertise for annotation and maintenance. Our folksonomy-based model allows associating semantic constraints on the input and output messages of web service operations using tags obtained from a folksonomy. We show how the model can be used for discovery and composition of services. We also describe a planner that uses this model to compose services and create workflows, automatically. We present performance results for the planner and our experiences in using this model in a sample real-world domain.

A semantics-based middleware for utilizing heterogeneous sensor networks

With the proliferation of various kinds of sensor networks, we will see large amounts of heterogeneous data. They have different characteristics such as data content, formats, modality and quality. Existing research has largely focused on issues related to individual sensor networks; how to make use of diverse data beyond the individual network level is largely unaddressed. In this paper, we propose a semantics-based approach for this problem and describe a system that constructs applications that utilize many sources of data simultaneously. We propose models to formally describe the semantics of data sources, and processing modules that perform various kinds of operations on data. Based on such formal semantics, our system composes data sources and processing modules together in response to users queries. The semantics provides a common ground such that data sources and processing modules from various parties can be shared and reused among applications. We describe our system architecture, illustrate application deployment, and share our experiences in the semantic approach.

Enhanced algorithm cost model to control tradeoffs in provisioning shared mesh restored lightpaths

In this write-up we propose an algorithm-centered metric to vary the weight put on the solutions cost and on the average backup lengths while selecting a primary-backup pair from a set of candidate routes. We assess the effect of our metric on these two contradicting objectives and show that it offers the leverage to achieve the desired compromise. We first present the cost model, we then describe the algorithm used in our experiments to illustrate the effect of this cost model, and we finally conclude with the results of our experiments.

Invited: Routing Strategies for Capacity-Efficient and Fast-Restorable Mesh Optical Networks

Wavelength division multiplexed (WDM)-based mesh network infrastructures that route optical connections using intelligent optical cross-connects (OXCs) are emerging as the technology of choice to implement the next generation core optical networks. In these architectures a single OXC is capable of switching tens of terabits of traffic per second. With such data transfer rates at stake, it becomes increasingly challenging for carriers to (1) efficiently and cost-effectively operate and manage their infrastructure, and (2) cope with network failures while guaranteeing prescribed service level agreements (SLAs) to their customers. Proper routing of primary and backup paths is a critical component of the routing and restoration architecture required to meeting these challenges. In this paper we review some of the various strategies and approaches proposed so far to intelligently route connections while at the same time providing guaranteed protection against various types of network failures. We explore the tradeoffs associated with these approaches, and investigate in particular different, sometimes competing aspects, such as cost/capacity required, level of protection (link vs. node failure), restoration time, and complexity of route computation.

Predicting arrival times of buses using real-time GPS measurements

Predicting arrival times of buses is a key challenge in the context of building intelligent public transportation systems. In this paper, we describe an efficient non-parametric algorithm which provides highly accurate predictions based on real-time GPS measurements. The key idea is to use a Kernel Regression model to represent the dependencies between position updates and the arrival times at bus stops. The performance of the proposed algorithm is evaluated on real data from the public bus transportation system in Dublin, Ireland. For a time horizon of 50 minutes, the prediction error of the algorithm is less than 10 percent on average. It clearly outperforms parametric methods which use a Linear Regression model, predictions based on the K-Nearest Neighbor algorithm, and a system which computes predictions of arrival times based on the current delay of buses. A study investigating the selection of interpolation points to reduce the size of the training set concludes the paper.