Jacques Carlier

Factoring and reductions for networks with imperfect vertices

Factoring and reductions are effective methods for computing the K-terminal reliability of undirected networks, but they have been applied mostly to networks with perfect vertices. However, in real problems, vertices may fail as well as edges. Imperfect vertices can be factored like edges, but the complexity then increases exponentially with their number. A technique has been developed to account for the failure of vertices with small additional cost, using a modified method of factoring and reductions. This technique is very easy to integrate into a factoring algorithm. It consists of factoring not on a single element (e.g., a single edge) but on a set of elements (e.g., an edge and its endpoints). The problem is that random variables associated with the elements of the network are no longer independent. This can be handled by choosing factoring edges that have at least one perfect endpoint. This technique leaves the factoring algorithm practically unchanged. The only difference is that some supplementary probability values are kept for the imperfect vertices of the original and the induced graphs. For algorithms using simple reductions, it has negligible computational cost. >

Modelling scheduling problems with timed Petri nets

In this paper, we show how to model with a timed Petri net, tasks, resources and constraints of a scheduling problem.

Using Linear Programming Methods for Optimizing the Real-Time Pump Scheduling

The work presented in this talk deals with the management of a drinking water distribution network in terms of planning the use of dier- ent installations (treatment works, pumping stations, and valves) in order to convey water from sources (rivers, borings, springs,...) to supply areas. More precisely we study the real-time pump scheduling problem. Being given a water distribution network and some previsions on the consumption at dierent nodes of the network during a considered time horizon, the main problem is to schedule water pump jobs under the constraint to satisfy water demands with the quality standards settled by French and European legisla- tion, while minimizing the operating costs (treatment and electricity). These operations should satisfy technical constraints as the respect of the minimum and maximum level of tanks, some contractual constraints as the respect of power levels defined by electricity supplier contracts, and last take some specific water distribution network constraints related to the impact of pressure in modeling or the need of the raw water flow to be as smooth as possible. The task is dicult

Assigning spare capacities in mesh survivable networks

This paper analyses one key issue of designing reliable networks: assignment of spare capacities in transmission networks. The spare capacities are optimized to facilitate the restoration of single failures. This problem can be formulated as an integer linear program and approximated by its continuous relaxation. This model is based on arc-path formulation especially efficient for dealing with end-to-end rerouting and providing appreciable economies in comparison with local rerouting. The main idea of our method resides in a linear programming decomposition, which permits us to solve problems for medium and large networks. Our approach could be applicable to both STM and ATM-based networks. This method was tested successfully on medium and large DCS-meshed networks and some numerical examples are given to illustrate its performances in terms of CPU time and ratio of optimality.

A GRASP for Placement and Routing of Dataflow Process Networks on Many-Core Architectures

We propose a GRASP heuristic for solving the joint problem of placement and routing of process networks from the field of compilation for embedded many core architectures. The method we propose consists in assigning applications expressed as dataflow process networks on homogeneous many core architectures by taking into account the routing maximal capacity of the arcs for the underlying Network-On-Chip. Our experiments illustrate the algorithm ability to efficiently obtain good quality routable assignments, within an acceptable computational time, even for large size instances. Moreover, validation of our approach is also realized on data coming from an embedded application of video processing.

A GRASP metaheuristic for the robust mapping and routing of dataflow process networks on manycore architectures

In this paper, we study the problem of joint placement and routing, both in the deterministic and stochastic cases, arising in the field of compilation of dataflow applications for manycore architectures. A GRASP algorithm is first proposed for solving the deterministic version and extended afterwards to treat the chance-constrained program with uncertainty affecting the weights of a dataflow process network. Extensive computational results, on representative synthetic benchmark and real data, illustrate the practical relevance of the approach, as well as the robustness of the obtained stochastic solutions.

Leveraging distributed GraphLab for program trace analysis

Graph-mining is a class of data-mining problems where programs involve the processing of data modeled as graphs. These applications often exhibit irregular and data-dependent communication patterns, hampering parallelization opportunities on distributed architectures. Many tools and frameworks were created for the scalable processing of graphs but their comparison is non-trivial on distributed architectures as there is no efficiency metrics with respect to distributed resource usage. Considering an in-house use-case, program trace analysis for parallelization optimizations, we study the benefits and limits of a graph-processing framework for a tangible application. The algorithm was implemented using GraphLab and executed on a humble 7-node commodity cluster with input instances up to 40 million vertices and 50 million edges. We propose in this paper an in-depth analysis of the GraphLab system to evaluate its performance and scalability in the context of program trace analysis. The analysis is driven both by traditional and domain-specific metrics and contributes to a better understanding of the system behavior.

Probabilistic Parameters of Conditional Task Graphs

This paper deals with the problem of determination of probabilistic parameters for tasks in a series-parallel conditional task graph. Such problematic is encountered in the context of parallel computing when dealing with conditional precedence constrained parallel tasks on a multi-core machine. The conditional task graph was introduced in order to express conditional precedence constraints and thus to model conditional execution in an application, which is not possible with a conventional task graph. We focus here in the calculation of two probabilistic parameters: the heads (release dates) and the tails (delivery times). An algorithm for computing these parameters is proposed. Although it has a pseudo-polynomial time complexity, the execution time of the algorithm can be further reduced at the price of less precision in the results.

Exploration of de Bruijn Graph Filtering for de novo Assembly Using GraphLab

The emergence of next generation DNA sequencers has raised interest in short read de novo assembly of whole genomes. Though numerous frameworks were developed in the field, the presence of errors in reads as well as the increasing size of datasets call for scalable preprocessing methods for noise filtering. In this paper we present a filtering algorithm that targets determination of valid k-mers in a de Bruijn graph built from short reads. Such preprocessing will help increase accuracy and reduce memory footprint in further assembly procedures by removing erroneous k-mers from the datasets at an early stage. The algorithm leverages GraphLab, a scalable graph processing framework not previously used in traditional assembly toolchains. The accuracy of the algorithm was evaluated with synthetic datasets exhibiting various error rates and proven to be able to determine large parts of de Bruijn graphs on datasets with error level greater than real-life datasets. The implementation is executed on a distributed cluster and a study of its scalability and operating performances is conducted and exhibits interesting scaling properties, hence demonstrating the relevance of GraphLab in such a context.

Distributed Rerouting in DCS Mesh Networks

Future telecommunication services will demand a fault-tolerant network with complete survivability. In this context, the reconfiguration of a network in real time has become one of the key issues on network reliability. In continuation of our previous work presented in [3], this paper proposes a fast distributed rerouting algorithm based on a new approach for generation of the restoration paths. The performances of the proposed algorithms are evaluated in terms of restoration efficiency and restoration time.