All publications sorted by year |
2024 |
Abstract: | As digital services are increasingly being deployed and used in a variety of domains, the environmental impact of Information and Communication Technologies (ICTs) is a matter of concern. Artificial intelligence is driving some of this growth but its environmental cost remains scarcely studied. A recent trend in large-scale generative models such as ChatGPT has especially drawn attention since their training requires intensive use of a massive number of specialized computing resources. The inference of those models is made accessible on the web as services, and using them additionally mobilizes end-user terminals, networks, and data centers. Therefore, those services contribute to global warming, worsen metal scarcity, and increase energy consumption. This work proposes an LCA-based methodology for a multi-criteria evaluation of the environmental impact of generative AI services, considering embodied and usage costs of all the resources required for training models, inferring from them, and hosting them online. We illustrate our methodology with Stable Diffusion as a service, an open-source text-to-image generative deep-learning model accessible online. This use case is based on an experimental observation of Stable Diffusion training and inference energy consumption. Through a sensitivity analysis, various scenarios estimating the influence of usage intensity on the impact sources are explored. |
2023 |
Abstract: | Service discovery is crucial in the development of fully decentralized computational grids. Among the significant amount of work produced by the convergence of peer- to-peer (P2P) systems and grids, a new kind of overlay networks, based on prefix trees (a.k.a., tries), has emerged. In particular, the Distributed Lexicographic Placement Table (DLPT) approach is a decentralized and dynamic service discovery service. Fault-tolerance within the DLPT approach is achieved through best-effort policies relying on formal self-stabilization results. Self-stabilization means that the tree can become transiently inconsistent, but is guaranteed to autonomously converge to a correct topology after arbitrary crashes, in a finite time. However, during convergence, the tree may not be able to process queries correctly. In this paper, we present some simula- tion results having several objectives. First, we investigate the interest of self-stabilization for such architectures. Second, we explore, still based on simulation, a simple Time-To-Live policy to avoid useless processing during convergence time. |
2022 |
Annotation: | https://emergingtechnet.org/SDN-NFV2022/ |
2021 |
2020 |
2019 |
2018 |
2017 |
Abstract: | The adoption of cloud computing is still limited by several legal concerns that customers may 68 have, such as data sovereignty. In cloud computing, data can be physically hosted in sensible loca- 69 tions, resulting in a lack of control for companies. In this context, we present the Nu@ge project, 70 which aims at building a federation of container-sized datacenters in the French territory. Nu@ge 71 provides a software stack that enables companies to interconnect independent datacenters in 72 a national mesh. A software architecture is presented and implemented as a federation of small 73 datacenters deployed in France. The proposed architecture enables cooperation between local 74 customized-cloud managers and a federation-wide middleware. It uses monitoring information 75 from facilities and performance indicators from physical servers for managing the system, pre- 76 venting incidents and considering energy efficiency. Additionally, a prototype of a container-sized 77 datacenter has been validated and patented. |
2016 |
2015 |
2014 |
2013 |
Abstract: | In the days of large scale, heterogeneous computing infrastructures gathering myriads of services, service discovery has become a critical feature that has to deal with the scale and dynamic nature of such platforms. The SPADES project (Servicing Petascale Architectures and DistributEd System, 08-ANR-SEGI-025) is a consortium whose purpose is to promote new solutions to deal with a very large number of volatile and heterogeneous computing resources. For the aforementioned reasons, at the project's core, the service discovery has been envisioned as fully decentralized. More precisely, the proposed P2P service discovery system proposed within the SPADES project is based on the DLPT approach (Distributed Lexicographic Placement Table) providing distributed structures and algorithms for such a feature. In this paper, an implementation of the DLPT concepts into the middleware developed within the SPADES project, called SBAM (Spades BAsed Middleware), is devised. Furthermore, its actual deployment over a nation-wide grid system, as well as its performance are detailed. |
Abstract: | With the increasing numbers of Cloud Service Providers and the migration of the Grids to the Cloud paradigm, it is necessary to be able to leverage these new resources. Moreover, a large class of High Performance Computing (HPC) applications can run these resources without (or with minor) modifications. But using these resources come with the cost of being able to interact with these new resource providers. In this paper we introduce the design of a HPC middleware that is able to use resources coming from an environment that compose of multiple Clouds as well as classical \hpc resources. Using the \diet middleware, we are able to deploy a large-scale, distributed HPC platform that spans across a large pool of resources aggregated from different providers. Furthermore, we hide to the end users the difficulty and complexity of selecting and using these new resources even when new Cloud Service Providers are added to the pool. Finally, we validate the architecture concept through cosmological simulation RAMSES. Thus we give a comparison of 2 well-known Cloud Computing Software: OpenStack and OpenNebula. |
2012 |
Abstract: | Infrastructure as a Service clouds are a flexible and fast way to obtain (virtual) resources as demand varies. Grids, on the other hand, are middleware platforms able to combine resources from different administrative domains for tasks ex- ecution. Clouds can be used as providers of devices such as virtual machines by grids so they only use the resources they need at every moment, but this requires grids to be able to decide when to allocate and release those resources. Here we analyze by simulations an economic approach to set resource prices and resolve when to scale resources depending on the users demand. Our simula- tor is based on the well-known GridSim software for grid simulation, which we expand. The results show how the proposed system can successfully adapt the amount of allocated resources to the demand, while at the same time ensuring that resources are fairly shared among users. |
Abstract: | Many scientific applications are described through workflow structures. Due to the increasing level of parallelism offered by modern computing infrastructures, workflow applications now have to be composed not only of sequential programs, but also of parallel ones. Cloud platforms bring on-demand resource provisioning and pay-as-you-go payment charging. Then the execution of a workflow corresponds to a certain budget. The current work addresses the problem of resource allocation for non-deterministic workflows under budget constraints. We present a way of transforming the initial problem into sub-problems that have been studied before. We propose two new allocation algorithms that are capable of determining resource allocations under budget constraints and we present ways of using them to address the problem at hand. |
Abstract: | Systems biology can now be considered an established and fundamental field in life sciences. It has facilitated the move from the identification of molecular `parts lists' for living organisms towards synthesising information from different `omics'-based approaches to generate and test new hypotheses about how biological systems work. This joint EMBL-EBI-Wellcome Trust course will combine lectures and led discussions to identify the key challenges, opportunities and bottlenecks, with practical sessions on network analysis and network-based modelling. |
2011 |
Abstract: | This paper surveys the risks brought by multitenancy in software platforms, along with the most prominent solutions proposed to address them. A multitenant platform hosts and executes software from several users (tenants). The platform must ensure that no malicious or faulty code from any tenant can interfere with the normal execution of other users? code or with the platform itself. This security requirement is specially relevant in Platform-as-a-Service (PaaS) clouds. PaaS clouds offer an execution environment based on some software platform. Unless PaaS systems are deemed as safe environments users will be reluctant to trust them to run any relevant application. This requires to take into account how multitenancy is handled by the software platform used as the basis of the PaaS offer. This survey focuses on two technologies that are or will be the platform-of-choice in many PaaS clouds: Java and .NET. We describe the security mechanisms they provide, study their limitations as multitenant platforms and analyze the research works that try to solve those limitations. We include in this analysis some standard container technologies (such as Enterprise Java Beans) that can be used to standardize the hosting environment of PaaS clouds. Also we include a brief discussion of Operating Systems (OSs) traditional security capacities and why OSs are unlikely to be chosen as the basis of PaaS offers. Finally, we describe some research initiatives that reinforce security by monitoring the execution of untrusted code, whose results can be of interest in multitenant systems. |
Abstract: | Infrastructure as a Service clouds are a flexible and fast way to obtain (virtual) resources as demand varies. Grids, on the other hand, are middleware platforms able to combine resources from different administrative domains for tasks ex- ecution. Clouds can be used as providers of devices such as virtual machines by grids so they only use the resources they need at every moment, but this requires grids to be able to decide when to allocate and release those resources. Here we analyze by simulations an economic approach to set resource prices and resolve when to scale resources depending on the users demand. Our simula- tor is based on the well-known GridSim software for grid simulation, which we expand. The results show how the proposed system can successfully adapt the amount of allocated resources to the demand, while at the same time ensuring that resources are fairly shared among users. |
Abstract: | This paper surveys the risks brought by multitenancy in software platforms, along with the most prominent solutions proposed to address them. A multitenant platform hosts and executes software from several users (tenants). The platform must ensure that no malicious or faulty code from any tenant can interfere with the normal execution of other users? code or with the platform itself. This security requirement is specially relevant in Platform-as-a-Service (PaaS) clouds. PaaS clouds offer an execution environment based on some software platform. Unless PaaS systems are deemed as safe environments users will be reluctant to trust them to run any relevant application. This requires to take into account how multitenancy is handled by the software platform used as the basis of the PaaS offer. This survey focuses on two technologies that are or will be the platform-of-choice in many PaaS clouds: Java and .NET. We describe the security mechanisms they provide, study their limitations as multitenant platforms and analyze the research works that try to solve those limitations. We include in this analysis some standard container technologies (such as Enterprise Java Beans) that can be used to standardize the hosting environment of PaaS clouds. Also we include a brief discussion of Operating Systems (OSs) traditional security capacities and why OSs are unlikely to be chosen as the basis of PaaS offers. Finally, we describe some research initiatives that reinforce security by monitoring the execution of untrusted code, whose results can be of interest in multitenant systems. |
2010 |
Abstract: | Service discovery is crucial in the development of fully decentralized computational grids. Among the significant amount of work produced by the convergence of peer- to-peer (P2P) systems and grids, a new kind of overlay networks, based on prefix trees (a.k.a., tries), has emerged. In particular, the Distributed Lexicographic Placement Table (DLPT) approach is a decentralized and dynamic service discovery service. Fault-tolerance within the DLPT approach is achieved through best-effort policies relying on formal self-stabilization results. Self-stabilization means that the tree can become transiently inconsistent, but is guaranteed to autonomously converge to a correct topology after arbitrary crashes, in a finite time. However, during convergence, the tree may not be able to process queries correctly. In this paper, we present some simula- tion results having several objectives. First, we investigate the interest of self-stabilization for such architectures. Second, we explore, still based on simulation, a simple Time-To-Live policy to avoid useless processing during convergence time. |
Abstract: | Thanks to the availability of Grids and their middleware, a seamless access to computation and storage resource is provided to application developers and scientists. The D{\'e}crypthon project is one example of such high performance platform. In this paper, we present the architecture of the platform, the middleware developed to ease the access to several servers deployed over France and an example of an application taking advantage of its capacity. |
Annotation: | Acceptance rate: 35.1 Acceptance rate: 35.1% 256 submitted papers, 90 acceptedubmitted papers, 90 accepted |
Abstract: | Service discovery is crucial in the development of fully decentralized computational grids. Among the significant amount of work produced by the convergence of peer- to-peer (P2P) systems and grids, a new kind of overlay networks, based on prefix trees (a.k.a., tries), has emerged. In particular, the Distributed Lexicographic Placement Table (DLPT) approach is a decentralized and dynamic service discovery service. Fault-tolerance within the DLPT approach is achieved through best-effort policies relying on formal self-stabilization results. Self-stabilization means that the tree can become transiently inconsistent, but is guaranteed to autonomously converge to a correct topology after arbitrary crashes, in a finite time. However, during convergence, the tree may not be able to process queries correctly. In this paper, we present some simula- tion results having several objectives. First, we investigate the interest of self-stabilization for such architectures. Second, we explore, still based on simulation, a simple Time-To-Live policy to avoid useless processing during convergence time. |
Abstract: | The Cloud phenomenon brings along the cost-saving benefit of dynamic scaling. Knowledge in advance is necessary as the virtual resources that Cloud computing uses have a setup time that is not negligible. We propose a new approach to the problem of workload prediction based on identifying similar past occurrences to the current short-term workload history. We present in detail the auto-scaling algorithm that uses the above approach as well as experimental results by using real-world data and an overall evaluation of this approach, its potential and usefulness. |
2009 |
Abstract: | The work presented in this paper aims at restricting the input parameter values of the semi-analytical model used in Galics and MoMaF, so as to derive which parameters influence the most the results, e.g., star forma- tion, feedback and halo recycling efficiencies, etc. Our approach is to proceed empirically: we run lots of simulations and derive the correct ranges of values. The computation time needed is so large, that we need to run on a grid of com- puters. Hence, we model Galics and MoMaF execution time and output files size, and run the simulation using a grid middleware: Diet. All the complexity of accessing resources, scheduling simulations and managing data is harnessed by Diet and hidden behind a web portal accessible to the users. |
Abstract: | The Cloud phenomenon is quickly growing towards becoming the de facto standard of Internet Computing, storage and hosting both in industry and academia. The large scalability possibilities offered by Cloud platforms can be harnessed not only for services and applications hosting but also as a raw on-demand computing resource. This paper proposes the use of a Cloud system as a raw computational on-demand resource for a Grid middleware. We illustrate a proof of concept by considering the DIET-Solve Grid middleware and the Eucalyptus open-source Cloud platform. |
Abstract: | The work presented in this paper aims at restricting the input parameter values of the semi-analytical model used in Galics and MoMaF, so as to derive which parameters influence the most the results, e.g., star forma- tion, feedback and halo recycling efficiencies, etc. Our approach is to proceed empirically: we run lots of simulations and derive the correct ranges of values. The computation time needed is so large, that we need to run on a grid of com- puters. Hence, we model Galics and MoMaF execution time and output files size, and run the simulation using a grid middleware: Diet. All the complexity of accessing resources, scheduling simulations and managing data is harnessed by Diet and hidden behind a web portal accessible to the users. |
Abstract: | {M}obile ad hoc networks as well as grid platforms are distributed, changing, and error prone environments. {C}ommunication costs within such infrastructure can be improved, or at least bounded, by using k-clustering. {A} k-clustering of a graph, is a partition of the nodes into disjoint sets, called clusters, in which every node is distance at most k from a designated node in its cluster, called the clusterhead. {A} self-stabilizing asynchronous distributed algorithm is given for constructing a k-clustering of a connected network of processes with unique {ID}s and weighted edges. {T}he algorithm is comparison-based, takes {O}(nk) time, and uses {O}(log n + log k) space per process, where n is the size of the network. {T}his is the first distributed solution to the k-clustering problem on weighted graphs. |
Abstract: | {M}obile ad hoc networks as well as grid platforms are distributed, changing, and error prone environments. {C}ommunication costs within such infrastructure can be improved, or at least bounded, by using k-clustering. {A} k-clustering of a graph, is a partition of the nodes into disjoint sets, called clusters, in which every node is distance at most k from a designated node in its cluster, called the clusterhead. {A} self-stabilizing asynchronous distributed algorithm is given for constructing a k-clustering of a connected network of processes with unique {ID}s and weighted edges. {T}he algorithm is comparison-based, takes {O}(nk) time, and uses {O}(log n + log k) space per process, where n is the size of the network. {T}his is the first distributed solution to the k-clustering problem on weighted graphs. |
Abstract: | The Cloud phenomenon is quickly growing towards becoming the de facto standard of Internet Computing, storage and hosting both in industry and academia. The large scalability possibilities offered by Cloud platforms can be harnessed not only for services and applications hosting but also as a raw on-demand computing resource. This paper proposes the use of a Cloud system as a raw computational on-demand resource for a Grid middleware. We illustrate a proof of concept by considering the DIET-Solve Grid middleware and the Eucalyptus open-source Cloud platform. |
2008 |
Abstract: | Grid middleware are the link between large scale (and distributed) platforms and applications. Managing such a software system and the grid environment itself can be a hard task when no dedicated (and integrated) tool exist. Some can be used through nice graphical interfaces, but they are usually dedicated to one or some limited tasks. They do not fulfill all the needs of a grid end-user who wants to deploy grid applications easily and rapidly. The aim of this paper is to present the case study of an all-in-one software system, designed for the management of a grid middleware and gathering user-friendly graphical interfaces answering to the various needs of end-users. Moreover the software system eases the use of the grid by avoiding the scripting layer under a nice GUI enabling the user a faster and more efficient use of the grid environment. By this means we demonstrate how the \ddb fulfills all the needs of a unified tool for grid management. This paper gives a comparison with existing and well-known tools dedicated to some specific tasks such as grid resources management, grid monitoring or middleware management. |
Abstract: | The efficiency of service discovery is a crucial point in the development of fully decentralized middlewares intended to manage large scale computational grids. The work conducted on this issue led to the design of many peer-to-peer fashioned approaches. More specifically, the need for flexibility and complexity in the service discovery has seen the emergence of a new kind of overlays, based on tries, also known as lexicographic trees. Although these overlays are efficient and well designed, they require a costly maintenance and do not accurately take into account the heterogeneity of nodes and the changing popularity of the services requested by users. In this paper, we focus on reducing the cost of the maintenance of a particular architecture, based on a dynamic prefix tree, while enhancing it with some load balancing techniques that dynamically adapt the load of the nodes in order to maximize the throughput of the system. The algorithms developed couple a self-organizing prefix tree overlay with load balancing techniques inspired by similar previous works undertaken for distributed hash tables. After some simulation results showing how our load balancing heuristics perform in such an overlay and compare to other heuristics, we provide a fair comparison of this architecture and similar overlays recently proposed. |
Abstract: | In this report, we tackle the problem of scheduling an Ocean-Atmosphere ap- plication used for climate prediction on the grid. An experiment is composed of several 1D-meshes of identical DAGs composed of parallel tasks. To obtain a good completion time, we divide groups of processors into sets each working on parallel tasks. The group sizes are chosen by computing the best makespan for several grouping possibilities. We improved this heuristic method by different means. The improvement yielding to the best makespan is the representation of the problem as an instance of the Knapsack problem. As this heuristic is firstly designed for homogeneous platforms, we present its adaptation to heterogeneous platforms. Simulations show improvements of the makespan up to 12 |
Abstract: | The use of many distributed, heterogeneous resources as a large collective platform offers great potential. A key issue for these grid platforms is middleware scalability and how middleware services can be mapped on the available resources. Optimizing deployment is a difficult problem with no existing general solutions. In this paper, we address the following problem: how to perform out an adapted deployment for a hierarchy of servers and resource brokers on a heterogeneous system? Our objective is to generate a best platform from the available nodes so as to fulfill the clients demands. However, finding the best deployment among heterogeneous resources is a hard problem since it is close to find the best broadcast tree in a general graph, which is known to be NP-complete. Thus, in this paper, we present a heuristic for middleware deployment on heterogeneous resources. We apply our heuristic to automatically deploy a distributed Problem Solving Environment on a large scale grid. We present experiments comparing the automatically generated deployment against a number of other reasonable deployments. |
Abstract: | The use of many distributed, heterogeneous resources as a large collective platform offers great potential. A key issue for these grid platforms is middleware scalability and how middleware services can be mapped on the available resources. Optimizing deployment is a difficult problem with no existing general solutions. In this paper, we address the following problem: how to perform out an adapted deployment for a hierarchy of servers and resource brokers on a heterogeneous system? Our objective is to generate a best platform from the available nodes so as to fulfill the clients demands. However, finding the best deployment among heterogeneous resources is a hard problem since it is close to find the best broadcast tree in a general graph, which is known to be NP-complete. Thus, in this paper, we present a heuristic for middleware deployment on heterogeneous resources. We apply our heuristic to automatically deploy a distributed Problem Solving Environment on a large scale grid. We present experiments comparing the automatically generated deployment against a number of other reasonable deployments. |
Abstract: | Grid middleware are the link between large scale (and distributed) platforms and applications. Managing such a software system and the grid environment itself can be a hard task when no dedicated (and integrated) tool exist. Some can be used through nice graphical interfaces, but they are usually dedicated to one or some limited tasks. They do not fulfill all the needs of a grid end-user who wants to deploy grid applications easily and rapidly. The aim of this paper is to present the case study of an all-in-one software system, designed for the management of a grid middleware and gathering user-friendly graphical interfaces answering to the various needs of end-users. Moreover the software system eases the use of the grid by avoiding the scripting layer under a nice GUI enabling the user a faster and more efficient use of the grid environment. By this means we demonstrate how the \ddb fulfills all the needs of a unified tool for grid management. This paper gives a comparison with existing and well-known tools dedicated to some specific tasks such as grid resources management, grid monitoring or middleware management. |
Abstract: | Grid middleware are the link between large scale (and distributed) platforms and applications. Managing such a software system and the grid environment itself can be a hard task when no dedicated (and integrated) tool exist. Some can be used through nice graphical interfaces, but they are usually dedicated to one or some limited tasks. They do not fulfill all the needs of a grid end-user who wants to deploy grid applications easily and rapidly. The aim of this paper is to present the case study of an all-in-one software system, designed for the management of a grid middleware and gathering user-friendly graphical interfaces answering to the various needs of end-users. Moreover the software system eases the use of the grid by avoiding the scripting layer under a nice GUI enabling the user a faster and more efficient use of the grid environment. By this means we demonstrate how the \ddb fulfills all the needs of a unified tool for grid management. This paper gives a comparison with existing and well-known tools dedicated to some specific tasks such as grid resources management, grid monitoring or middleware management. |
Abstract: | The efficiency of service discovery is a crucial point in the development of fully decentralized middlewares intended to manage large scale computational grids. The work conducted on this issue led to the design of many peer-to-peer fashioned approaches. More specifically, the need for flexibility and complexity in the service discovery has seen the emergence of a new kind of overlays, based on tries, also known as lexicographic trees. Although these overlays are efficient and well designed, they require a costly maintenance and do not accurately take into account the heterogeneity of nodes and the changing popularity of the services requested by users. In this paper, we focus on reducing the cost of the maintenance of a particular architecture, based on a dynamic prefix tree, while enhancing it with some load balancing techniques that dynamically adapt the load of the nodes in order to maximize the throughput of the system. The algorithms developed couple a self-organizing prefix tree overlay with load balancing techniques inspired by similar previous works undertaken for distributed hash tables. After some simulation results showing how our load balancing heuristics perform in such an overlay and compare to other heuristics, we provide a fair comparison of this architecture and similar overlays recently proposed. |
Abstract: | The efficiency of service discovery is a crucial point in the development of fully decentralized middlewares intended to manage large scale computational grids. The work conducted on this issue led to the design of many peer-to-peer fashioned approaches. More specifically, the need for flexibility and complexity in the service discovery has seen the emergence of a new kind of overlays, based on tries, also known as lexicographic trees. Although these overlays are efficient and well designed, they require a costly maintenance and do not accurately take into account the heterogeneity of nodes and the changing popularity of the services requested by users. In this paper, we focus on reducing the cost of the maintenance of a particular architecture, based on a dynamic prefix tree, while enhancing it with some load balancing techniques that dynamically adapt the load of the nodes in order to maximize the throughput of the system. The algorithms developed couple a self-organizing prefix tree overlay with load balancing techniques inspired by similar previous works undertaken for distributed hash tables. After some simulation results showing how our load balancing heuristics perform in such an overlay and compare to other heuristics, we provide a fair comparison of this architecture and similar overlays recently proposed. |
2007 |
Abstract: | Within computational Grids, some services (typically software components, e.g., linear algebra libraries) are made available by some servers to some clients. In spite of the growing popularity of such Grids, the service discovery, although efficient in many cases, does not reach several requirements. Among them, the flexibility of the discovery and its efficiency on wide-area dynamic platforms are two major issues. Therefore, it becomes crucial to propose new tools coping with such platforms. Emerging peer-to-peer technologies provide algorithms allowing the distribution and the retrieval of data items while addressing the dynamicity of the underlying network. Whereas merging peer-to-peer technology and Grid infrastructures has been widely suggested, very few implementations are available. The contribution of this paper is twofold. First, we present the design, the implementation and the experimentation of the first architecture, to our knowledge, extending traditional Network-Enabled Servers (NES) systems with an unstructured peer-to-peer network. This extension allows to dynamically connect distributed agents thus providing to clients an entry point to servers geographically distributed. Our implementation is based on the Diet middleware and the JXTA toolbox and experimentation have been conducted on a high speed network. Then, we study the service discovery in a pure peer-to-peer environment. We describe a new trie-based approach for the peer-to-peer service discovery service, supporting range queries while providing fault-tolerance and taking into account the topology of the underlying network. We validate this approach both by analysis and simulation. |
Abstract: | In this paper, we study and compare grid and global computing systems and outline the benefits of having a hybrid system called DIRAC. To evaluate the DIRAC scheduling for high throughput computing, a new model is presented and a simulator was developed for many clusters of heterogeneous nodes belonging to a local network. These clusters are assumed to be connected to each other through a global network and each cluster is managed via a local scheduler which is shared by many users. We validate our simulator by comparing the experimental and analytical results of a M/M/4 queuing system. Next, we do the comparison with a real batch system and we obtain an average error of 10.5 0.000000or the response time and 12 0.000000or the makespan. We conclude that the simulator is realistic and well describes the behaviour of a large-scale system. Thus we can study the scheduling of our system called DIRAC in a high throughput context. We justify our decentralized, adaptive and opportunistic approach in comparison to a centralized approach in such a context. |
Abstract: | Within computational grids, some services (software components, linear algebra libraries, etc.) are made available by some servers to some clients. In spite of the growing popularity of such grids, the service discovery, although efficient in many cases, does not reach several requirements. Among them, the flexibility of the discovery and its efficiency on wide-area dynamic platforms are two major issues. Therefore, it becomes crucial to propose new tools coping with such platforms. Emerging peer-to-peer technologies provide algorithms allowing the distribution and the retrieval of data items while addressing the dynamicity of the underlying network. We study in this paper the service discovery in a pure peer-to-peer environment. We describe a new trie-based approach for the service discovery that supports range queries and automatic completion of partial search strings, while providing fault-tolerance, and partially taking into account the topology of the underlying network. We validate this approach both by analysis and simulation. Traditional metrics considered in peer-to-peer systems exhibits interesting complexities within our architecture. The analysis' results are confirmed by some simulation experiments run using several grid's data sets. |
Abstract: | Within computational grids, some services (software components, linear algebra libraries, etc.) are made available by some servers to some clients. In spite of the growing popularity of such grids, the service discovery, although efficient in many cases, does not reach several requirements. Among them, the flexibility of the discovery and its efficiency on wide-area dynamic platforms are two major issues. Therefore, it becomes crucial to propose new tools coping with such platforms. Emerging peer-to-peer technologies provide algorithms allowing the distribution and the retrieval of data items while addressing the dynamicity of the underlying network. We study in this paper the service discovery in a pure peer-to-peer environment. We describe a new trie-based approach for the service discovery that supports range queries and automatic completion of partial search strings, while providing fault-tolerance, and partially taking into account the topology of the underlying network. We validate this approach both by analysis and simulation. Traditional metrics considered in peer-to-peer systems exhibits interesting complexities within our architecture. The analysis' results are confirmed by some simulation experiments run using several grid's data sets. |
Abstract: | Resource Discovery is a crucial issue in the deployment of computational grids over large scale peer-to-peer platforms. Because they efficiently allow range queries, Prefix Trees appear to be among promising ways in the design of distributed data structures indexing resources. Self-stabilization is an efficient approach in the design of reliable solutions for dynamic systems. A snap-stabilizing algorithm guarantees that it always behaves according to its specification. In other words, a snap-stabilizing algorithm is also a self-stabilizing algorithm which stabilizes in 0 steps. In this paper, we provide the first snap-stabilizing protocol for trie construction. We design particular tries called Proper Greatest Common Prefix (PGCP) Tree. The proposed algorithm arranges the n label values stored in the tree, in average, in O(h+h') rounds, where h and h' are the initial and final heights of the tree, respectively. In the worst case, the algorithm requires an O(n) extra space on each node, O(n) rounds and O(n^2) actions. However, simulations show that, using relevant data sets, this worst case is far from being reached and confirm the average complexities, making this algorithm efficient in practice. |
2006 |
Abstract: | Among existing grid middleware approaches, one simple, powerful, and flexible approach consists of using servers available in different administrative domains through the classical client-server or Remote Procedure Call (RPC) paradigm. Network Enabled Servers implement this model also called GridRPC. Clients submit computation requests to a scheduler whose goal is to find a server available on the grid. The aim of this paper is to give an overview of a middleware developed by the GRAAL team called DIET (for Distributed Interactive Engineering Tool-box). DIET is a hierarchical set of components used for the development of applications based on computational servers on the grid. |
Abstract: | Among existing grid middleware approaches, one simple, powerful, and flexible approach consists of using servers available in different administrative domains through the classic client-server or Remote Procedure Call (RPC) paradigm. Network Enabled Servers (NES) implement this model also called GridRPC. Clients submit computation requests to a scheduler whose goal is to find a server available on the grid. The aim of this paper is to give an overview of an NES middleware developed in the GRAAL team called DIET and to describe recent developments. DIET (Distributed Interactive Engineering Toolbox) is a hierarchical set of components used for the development of applications based on computational servers on the grid. |
Abstract: | Among existing grid middleware approaches, one simple, powerful, and flexible approach consists of using servers available in different administrative domains through the classic client-server or Remote Procedure Call (RPC) paradigm. Network Enabled Servers (NES) implement this model also called GridRPC. Clients submit computation requests to a scheduler whose goal is to find a server available on the grid. The aim of this paper is to give an overview of an NES middleware developed in the GRAAL team called DIET and to describe recent developments. DIET (Distributed Interactive Engineering Toolbox) is a hierarchical set of components used for the development of applications based on computational servers on the grid. |
Abstract: | Among existing grid middleware approaches, one simple, powerful, and flexible approach consists of using servers available in different administrative domains through the classic client-server or Remote Procedure Call (RPC) paradigm. Network Enabled Servers (NES) implement this model also called GridRPC. Clients submit computation requests to a scheduler whose goal is to find a server available on the grid. The aim of this paper is to give an overview of an NES middleware developed in the GRAAL team called DIET and to describe recent developments. DIET (Distributed Interactive Engineering Toolbox) is a hierarchical set of components used for the development of applications based on computational servers on the grid. |
Abstract: | As grids become more and more attractive for solving complex problems with high computational and storage requirements, the need for adequate grid programming models is considerable. To this purpose, the GridRPC model has been proposed as a grid version of the classical RPC paradigm, with the goal to build NES (Network-Enabled Server) environments. Paradoxically enough, in this model, data management has not been defined and is now explicitly left at the user's charge. The contribution of this paper is to enhance data management in NES by introducing a transparent data access model, available through the concept of grid data-sharing service. Data management (persistent storage, transfer, consistent replication) is totally delegated to the service, whereas the applications simply access shared data via global identifiers. We illustrate our approach using the DIET GridRPC middleware and the JUXMEM data-sharing service. Experiments performed on the Grid'5000 testbed demonstrate the benefits of the proposed approach. |
Abstract: | As grids become more and more attractive for solving complex problems with high computational and storage requirements, the need for adequate grid programming models is considerable. To this purpose, the GridRPC model has been proposed as a grid version of the classical RPC paradigm, with the goal to build NES (Network-Enabled Server) environments. Paradoxically enough, in this model, data management has not been defined and is now explicitly left at the user's charge. The contribution of this paper is to enhance data management in NES by introducing a transparent data access model, available through the concept of grid data-sharing service. Data management (persistent storage, transfer, consistent replication) is totally delegated to the service, whereas the applications simply access shared data via global identifiers. We illustrate our approach using the DIET GridRPC middleware and the JUXMEM data-sharing service. Experiments performed on the Grid'5000 testbed demonstrate the benefits of the proposed approach. |
Abstract: | One of the true challenges in resource management in grids is to provide support for co-allocation, that is, the allocation of resources in multiples autonomous subsystems of a grid to single jobs. With reservation-based local schedulers, a grid scheduler can reserve processors with these schedulers to achieve simultaneous processor availability. However, with queuing-based local schedulers, it is much more difficult to guarantee this. In this paper we present mechanisms and policies for working around the lack of reservation mechanisms for jobs with deadlines that require co-allocation, and simulations of these mechanisms and policies. |
Abstract: | This report presents the approach chosen within the DIET (Distributed Interactive Engineering Toolbox) project a GridRPC environment to allow a resource broker to be tuned for specific application classes. Our design allows the use of generic or application dependent performance measures in a simple and seamless way. |
Abstract: | This report presents the approach chosen within the DIET (Distributed Interactive Engineering Toolbox) project a GridRPC environment to allow a resource broker to be tuned for specific application classes. Our design allows the use of generic or application dependent performance measures in a simple and seamless way. |
Abstract: | Facing the limits of traditional tools of resource management within computational grids (related to scale, dynamicity, etc. of the platforms newly considered), new approaches, based on peer-to-peer technologies are emerging. The resource discovery and in particular the service discovery is concerned by this evolution. Among the solutions, a promising one is the indexing of resources using trie structures and more particularly prefix trees. The major advantages of trie-structured approaches is the capability to support search queries on ranges of values with a latency growing logarithmically in the number of nodes in the trie. Those techniques are easy to extend to multicriteria searches. One drawback of using tries is its inherent poor robustness in a dynamic environment, where nodes join and leave the network, leading to the split of the tree into a forest, which results in the impossibility to route requests. Within most recent approaches, the fault-tolerance is a prevention mechanism, often replication-based. The replication can be costly in term of resources required. In this paper, we propose a fault-tolerance protocol that reconnects subtrees a posteriori, after crashes, to have again a connected graph and then reorder the nodes to rebuild a consistent tree. |
Abstract: | Facing the limits of traditional tools of resource management within computational grids (related to scale, dynamicity, etc. of the platforms newly considered), new approaches, based on peer-to-peer technologies are emerging. The resource discovery and in particular the service discovery is concerned by this evolution. Among the solutions, a promising one is the indexing of resources using trie structures and more particularly prefix trees. The major advantages of trie-structured approaches is the capability to support search queries on ranges of values with a latency growing logarithmically in the number of nodes in the trie. Those techniques are easy to extend to multicriteria searches. One drawback of using tries is its inherent poor robustness in a dynamic environment, where nodes join and leave the network, leading to the split of the tree into a forest, which results in the impossibility to route requests. Within most recent approaches, the fault-tolerance is a prevention mechanism, often replication-based. The replication can be costly in term of resources required. In this paper, we propose a fault-tolerance protocol that reconnects subtrees a posteriori, after crashes, to have again a connected graph and then reorder the nodes to rebuild a consistent tree. |
Abstract: | Within computational grids, some services (software components, linear algebra libraries, etc.) are made available by some servers to some clients. In spite of the growing popularity of such grids, the service discovery, although efficient in many cases, does not reach several requirements. Among them, the flexibility of the discovery and its efficiency on wide-area dynamic platforms are two major issues. Therefore, it becomes crucial to propose new tools coping with such platforms. Emerging peer-to-peer technologies provide algorithms allowing the distribution and the retrieval of data items while addressing the dynamicity of the underlying network. We study in this paper the service discovery in a pure peer-to-peer environment. We describe a new trie-based approach for the service discovery that supports range queries and automatic completion of partial search strings, while providing fault-tolerance, and partially taking into account the topology of the underlying network. We validate this approach both by analysis and simulation. Traditional metrics considered in peer-to-peer systems exhibits interesting complexities within our architecture. The analysis' results are confirmed by some simulation experiments run using several grid's data sets. |
Abstract: | Within computational grids, some services (software components, linear algebra libraries, etc.) are made available by some servers to some clients. In spite of the growing popularity of such grids, the service discovery, although efficient in many cases, does not reach several requirements. Among them, the flexibility of the discovery and its efficiency on wide-area dynamic platforms are two major issues. Therefore, it becomes crucial to propose new tools coping with such platforms. Emerging peer-to-peer technologies provide algorithms allowing the distribution and the retrieval of data items while addressing the dynamicity of the underlying network. We study in this paper the service discovery in a pure peer-to-peer environment. We describe a new trie-based approach for the service discovery that supports range queries and automatic completion of partial search strings, while providing fault-tolerance, and partially taking into account the topology of the underlying network. We validate this approach both by analysis and simulation. Traditional metrics considered in peer-to-peer systems exhibits interesting complexities within our architecture. The analysis' results are confirmed by some simulation experiments run using several grid's data sets. |
2005 |
2004 |
2003 |
Abstract: | This paper focus on the deployment of grid infrastructures, more specifically Problem Solving Environments (PSE) for numerical applications on the grid. Even if the deployment of such an architecture is forced by physical constraints (firewall, access permission, security,...) its efficiency heavily depends on the quality of the mapping between its different components and the grid resources. This paper proposes a new model based on linear programming to estimate the performance of a deployment of a hierarchical PSE. The advantages of the modeling approach in this case are multiple: evaluate a virtual deployment before an actual deployment, provide a decision builder tool (i.e., designed to compare different architectures or buy new resource), take into account the platform scalability. Using this modeling, it is possible to determine the bottleneck of the platform and thus to know whether a given deployment can be improved or not. We illustrate this modeling by applying this results to an existing hierarchical PSE called DIET. |
2002 |
2001 |
Abstract: | In this paper, we present the developments realized in the OURAGAN project around the parallelization of a MATLAB-like tool called SCILAB. These developments use high-performance numerical libraries and different approaches based either on the duplication of SCILAB processes or on computational servers. This tool, SCILAB//, allows users to perform high-level operations on distributed matrices in a metacomputing environment. We also present performance results on different architectures. |
2000 |
Abstract: | La recherche en parall{\'e}lisme s'est concentr{\'e}e avec succ{\`e}s sur les aspects calcul et communication, on peut aujourd'hui pr{\'e}tendre {\`a} des puissances de calcul de l'ordre du T{\'e}ra-flops (milliard d'op{\'e}ration par seconde). Les applications qui requi{\`e}rent de telle puissance de calcul sont en g{\'e}n{\'e}rales celles qui ont {\`a} traiter des masses de donn{\'e}es qui se mesurent en Giga-octects, voir en T{\'e}ra-octets. Or, dans la pratique les capacit{\'e}s m{\'e}moire fixent la taille maximale du probl{\`e}me que l'utilisateur pourra traiter. L'utilisateur consid{\'e}rera qu'il est inconcevable de recourir au syst{\`e}me de pagination ou aux disques du fait de l'importante chute des performances. Et pourtant, le paradoxe est l{\`a} d'un c{\^o}t{\'e} des capacit{\'e}s m{\'e}moire rapide mais de petite taille et co{\^u}teuses, de l'autre des capacit{\'e}s m{\'e}moire {\`a} bon march{\'e} et de grande taille, plusieurs T{\'e}ra-octets, mais tr{\`e}s lente d'acc{\`e}s. Le concept du calcul out-of-core r{\'e}ponds {\`a} ces attentes en proposant d'utiliser au mieux les ressources des m{\'e}moires externes. Cette th{\`e}se d{\'e}veloppe une {\'e}tude sur le calcul num{\'e}rique hautes performances out-of-core et propose des m{\'e}canismes de traitement des donn{\'e}es de tr{\`e}s grande taille. Pour r{\'e}soudre efficacement le probl{\`e}me de la gestion m{\'e}moire, deux approches diff{\'e}rentes sont trait{\'e}es. Une approche syst{\`e}me dans laquelle nous {\'e}tudions les m{\'e}canismes de pagination. Nous d{\'e}montrons que ces syst{\`e}mes ne sont pas adapt{\'e}s au calcul num{\'e}rique intensif, et qu'il est alors important de r{\'e}ordonnancer les flux d'ex{\'e}cution. Nous pr{\'e}sentons une nouvelle biblioth{\`e}que de gestion de la m{\'e}moire virtuelle au niveau utilisateur. Une optimisation de la factorisation LU de tr{\`e}s grandes matrices illustre l'utilisation de cette biblioth{\`e}que. La seconde approche est bas{\'e}e sur l'algorithmique out-of-core. En pr{\'e}ambule, nous pr{\'e}sentons la biblioth{\`e}que de calcul d'alg{\`e}bre lin{\'e}aire ScaLAPACK, en offrant une description de ses diff{\'e}rentes composantes. Un soin particulier est port{\'e} sur la pr{\'e}sentation du prototype out-of-core et les d{\'e}veloppements que nous avons effectu{\'e} au sein de ce prototype. Nous proposons, ensuite, une {\'e}tude d{\'e}taill{\'e}e de deux algorithmes out-of-core d{\'e}velopp{\'e}s dans cette biblioth{\`e}que : - la factorisation LU : nous pr{\'e}sentons un mod{\`e}le analytique permettant d'{\'e}valuer les performances de la factorisation LU parall{\`e}le out-of-core left-looking. L'objectif de cette mod{\'e}lisation est de d{\'e}celer des optimisations pour l'algorithme en question et de mettre en {\'e}vidence les surco{\^u}ts out-of-core. Le mod{\`e}le de pr{\'e}diction des performances est valid{\'e} de mani{\`e}re exp{\'e}rimentale. Nous prouverons ainsi qu'une distribution de la matrice correctement effectu{\'e}e et la mise en place d'un syst{\`e}me de recouvrement du surco{\^u}t des E/S par le calcul, il est possible d'obtenir des performances proches de l'algorithme en m{\'e}moire. De plus, nous d{\'e}terminons la taille m{\'e}moire minimale n{\'e}cessaire pour permettre ce recouvrement. - l'inversion matricielle : Nous proposons un algorithme d'inversion matricielle out-of-core par extension des travaux r{\'e}alis{\'e}s sur la factorisation LU out-of-core. Par extension du mod{\`e}le de pr{\'e}diction des performances de la factorisation LU, nous avons {\'e}labor{\'e} un mod{\`e}le analytique pour l'inversion. Ce mod{\`e}le permet de mettre en {\'e}vidence les surco{\^u}ts dus au traitement out-of-core. Nous montrons comment r{\'e}duire ces surco{\^u}ts afin d'obtenir une inversion out-of-core pouvant atteindre des performances proches d'une r{\'e}solution en m{\'e}moire. Afin d'exploiter ces r{\'e}sultats nous avons r{\'e}alis{\'e} pour ces deux algorithmes une version effectuant le recouvrement des acc{\`e}s disques par les temps de calcul. La technique utilis{\'e}e pour effectuer les recouvrements est bas{\'e}e sur le clonage de processus, apr{\`e}s avoir {\'e}voqu{\'e} le principe de mise en oeuvre, nous exposons les difficult{\'e}s rencontr{\'e}es pour obtenir un r{\'e}el recouvrement des entr{\'e}es/sorties. Par extension, nous pr{\'e}sentons alors la factorisation LU out-of-core et l'inversion matricielle out-of-core dans lesquelles nous avons mis en place le syst{\`e}me de recouvrement ainsi que les r{\'e}sultats obtenus. Une derni{\`e}re partie pr{\'e}sente l'int{\'e}gration de ces travaux dans Scilab//, un outil de calcul num{\'e}rique. La participation {\`a} ce projet de l'INRIA, permet d'utiliser la biblioth{\`e}que out-of-core de ScaLAPACK avec une interface conviviale, et, d'accro{\^\i}tre ainsi le champ d'application de Scilab//. Nous pr{\'e}sentons Scilab et les techniques permettant d'interfacer des modules externes compil{\'e}s. Nous pr{\'e}sentons ensuite Scilab//, et exposons l'extension que nous avons r{\'e}alis{\'e} afin de proposer une version out-of-core de Scilab. Nous menons ensuite une {\'e}tude afin de proposer des optimisations pour les traitements out-of-core. |
1999 |
1998 |
1997 |
1996 |
This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All person copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder.
Les documents contenus dans ces répertoires sont rendus disponibles par les auteurs qui y ont contribué en vue d'assurer la diffusion à temps de travaux savants et techniques sur une base non-commerciale. Les droits de copie et autres droits sont gardés par les auteurs et par les détenteurs du copyright, en dépit du fait qu'ils présentent ici leurs travaux sous forme électronique. Les personnes copiant ces informations doivent adhérer aux termes et contraintes couverts par le copyright de chaque auteur. Ces travaux ne peuvent pas être rendus disponibles ailleurs sans la permission explicite du détenteur du copyright.
This document was translated from BibTEX by bibtex2html