Search

International audience; ConspectusThe chemical study of materials from natural history and cultural heritage, which provide information for art history, archeology, or paleontology, presents a series of specific challenges. The complexity of these ancient and historical materials, which are chemically heterogeneous, the product of alteration processes, and inherently not reproducible, is a major obstacle to a thorough understanding of their making and long-term behavior (e.g., fossilization). These challenges required the development of methodologies and instruments coupling imaging and data processing approaches that are optimized for the specific properties of the materials. This Account discusses how these characteristics not only constrain their study but also open up specific innovative avenues for providing key historical information. Synchrotron methods have extensively been used since the late 1990s to study heritage objects, in particular for their potential to provide speciation information from excitation spectroscopies and to image complex heritage objects and samples in two and three dimensions at high resolution. We examine in practice how the identification of key intrinsic chemical specificities has offered fertile ground for the development of novel synchrotron approaches allowing a better stochastic description of the properties of ancient and historical materials. These developments encompass three main aspects: (1) The multiscale heterogeneity of these materials can provide an essential source of information in the development of probes targeting their multiple scales of homogeneity. (2) Chemical alteration can be described in many ways, e.g., by segmenting datasets in a semiquantitative way to jointly inform morphological and chemical transformation pathways. (3) The intrinsic individuality of chemical signatures in artifacts triggers the development of specific strategies, such as those focusing on weak signal detection. We propose a rereading of the advent of these new methodologies for analysis and characterization and examine how they have led to innovative strategies combining materials science, instrument development, history, and data science. In particular, we show that spectral imaging and the search for correlations in image datasets have provided a powerful way to address what archeologists have called the uncertainty and ambiguity of the material record. This approach has implications beyond synchrotron techniques and extends in particular to a series of rapidly developing approaches that couple spectral and spatial information, as in hyperspectral imaging and spatially resolved mass spectrometry. The preeminence of correlations holds promise for the future development of machine learning methods for processing data on historical objects. Beyond heritage, these developments are an original source of inspiration for the study of materials in many related fields, such as environmental, geochemical, or life sciences, which deal with systems whose alteration and heterogeneity cannot be neglected.

Recently, MapReduce and its open-source implementation Hadoop have emerged as prevalent tools for big data analysis in the cloud. Fair resource allocation in-between jobs and users is an important issue, especially in multi-tenant environments such as clouds. Thus several scheduling policies have been developed to preserve fairness in multi-tenant Hadoop clusters. At the core of these schedulers, simple (non-) preemptive approaches are employed to free resources for tasks belonging to jobs with less-share. For example, Hadoop Fair Scheduler is equipped with two approaches: wait and kill. While wait may introduce a serious violation in fairness, kill may result in a huge waste of resources. Yet, recently some works have introduced new preemption approaches (e.g., pause-resume) in shared Hadoop clusters. To this end, in this work, we closely examine three approaches including wait, kill and pause-resume when Hadoop Fair Scheduler is employed for ensuring fair execution between multiple concurrent jobs. We perform extensive experiments to assess the impact of these approaches on performance and resource utilization while ensuring fairness. Our experimental results bring out the differences between these approaches and illustrate that these approaches are only sub-optimal for different workloads and cluster configurations: the efficiency of achieving fairness and the overall performance varies with the workload composition, resource availability and the cost of the adopted preemption technique.; Récemment, le paradigme MapReduce et son implémentation open-source Hadoop sont devenus des outils très populaires pour l’analyse dedonnées massives dans le Cloud. Le partage équitable des ressources entre les différentes tâches et utilisateurs est un problème important, en particulier dans les architectures multi-tenant comme le Cloud. De nombreuses stratégies d’ordonnancement ont donc été développées pour préserver l’équité dans les cluster Hadoop multi-tenant. Au cœur de ces ordonnanceurs, des approches simples et non-préemptives sont utilisées pour libérer des ressources pour des tâches appartenant à des utilisateurs en ayant eu jusque-là une part plus faible. Par exemple, Hadoop Fair Scheduler possède deux approches : "attendre" et "tuer". Si "attendre" peut causer des sérieuses ruptures d’équité, "tuer" peut aussi entraîner un important gaspillage des ressources. Cependant, certains travaux récents ont introduit des techniques préemptives (c’est-à-dire "arrêter-reprendre") dans les clusters Hadoop partagés. Dans ce travail, nous examinons précisément trois approches, incluant "attendre", "tuer" et "arrêter-reprendre",lorsque Hadoop Fair Scheduler est utilisé pour assurer une répartition équitable des ressources lors de l’exécution de plusieurs groupes de tâches concurrents. Nous avons mené des expériences étendues pour évaluer l’impact de ces approches sur les performances et l’utilisation des ressources tout en garantissant leur partage équitable. Les résultats de nos expériences mettent en évidence les différences entre ces stratégies et montrent que chacune est sous-optimale pour une partie des workloads et des configurations : la capacité à garantir l’équité et les performances globales varient en fonction de la composition des tâches, des ressources disponibles et du coût des techniques préemptives.