This paper describes a distributed coordinated checkpointing protocol that always ensures a consistent set of checkpoints. A checkpoint initiator initiates checkpointing activity and the protocol followed is two phase...
详细信息
This paper describes a distributed coordinated checkpointing protocol that always ensures a consistent set of checkpoints. A checkpoint initiator initiates checkpointing activity and the protocol followed is two phase with each process maintaining a tentative checkpoint till it is made permanent or aborted. However, there is no central checkpoint initiator, but each of the processes takes turn to act as the initiator. Processes take local checkpoints only after being notified by the initiator. The guaranty that no message would be lost in the system (where processes communicate via messages only) in case of failure is maintained in this work by forcing processes to refrain from sending computation messages for a certain period of time that generally equals the time a message in a network takes to reach its destination from the sender. Processes carry out local computations only during that period that eventually gets included in the current permanent checkpoint.
Nodes in a distributed System are susceptible to failures for many different reasons. In case of such failures the distributed system as a whole needs to be restored to an error free state, existing prior to failure. ...
详细信息
Nodes in a distributed System are susceptible to failures for many different reasons. In case of such failures the distributed system as a whole needs to be restored to an error free state, existing prior to failure. This restoration is done by rolling back the computation at the nodes to an error free state. To minimize the amount of computation which needs to be rolled back checkpoints or snapshots of a globally consistent state are taken from time to time. We present a synchronous checkpointing algorithm which forces a minimum number of nodes to take a checkpoint. Underlying computation need not be blocked completely during the progress of the algorithm. No additional effort needs to be expended to circumvent the problem of concurrent initiations of the algorithm, since the initiator node assumes the responsibility of completing one instance before another one can be initiated. Since the consistency of the snapshots is ensured at the time the global snapshot is taken, no time needs to be spent during recovery.
It is often required to conduct data mining among several sites to discover valuable patterns, associations, trends, and dependencies in the shared data. The privacy, however, is a concern. In many situations users do...
详细信息
It is often required to conduct data mining among several sites to discover valuable patterns, associations, trends, and dependencies in the shared data. The privacy, however, is a concern. In many situations users do require that data mining is conducted with no privacy of any site being leaked out to any other sites. In this paper a distributed privacy-preserving data mining algorithm is proposed, which is characterized with its ability to resist the collusion, and especially, a system with more sites tends to have the ability to resist more collusion. Performance analysis results are provided for demonstrating the effectiveness of the proposed algorithm.
The Web has proved to be very efficient and effective in delivering business data to users everywhere. Using the Web to deliver business applications has been more challenging due to the limitations of connection perf...
详细信息
The Web has proved to be very efficient and effective in delivering business data to users everywhere. Using the Web to deliver business applications has been more challenging due to the limitations of connection performance and client interface capabilities. We present a new distributed architecture for developing network enterprise applications with highly interactive rich client interfaces that can be deployed effectively on the Internet. The components are executed on the server side, eliminates the need and delay for transporting application code over the network as well as the serious security issues associated with executing downloaded programs on client machines. The new architecture also provides an ideal solution for enabling convenient remote access to network-enabled devices from anywhere.
The present work describes a checkpointing protocol with a Mobile Support Station (MSS) acting as checkpoint initiator. The task of the initiator MSS is to issue checkpointing requests to all other MSSs. Each MSS then...
详细信息
The present work describes a checkpointing protocol with a Mobile Support Station (MSS) acting as checkpoint initiator. The task of the initiator MSS is to issue checkpointing requests to all other MSSs. Each MSS then forwards this request to all the Mobile Hosts (MHs) it is connected with. An MH may be inactive for a long time and hence its state has not changed (since it may have performed only local computation and not sent or received any computation message since its last checkpointing activity). Such an MH decides not to take a checkpoint thereby saving its power whereas other MHs and MSSs take checkpoint. An MSS however, has to save, besides its own state, all unacknowledged messages of each MHs that is currently registered with it.
In this paper, we present the improving capability of accuracy and the parallel efficiency of self-organizing neural groves (SONGs) for classification on a MIMD parallel computer. Self-generating neural networks (SGNN...
详细信息
In this paper, we present the improving capability of accuracy and the parallel efficiency of self-organizing neural groves (SONGs) for classification on a MIMD parallel computer. Self-generating neural networks (SGNNs) are originally proposed on adopting to classification or clustering by automatically constructing self-generating neural tree (SGNT) from given training data. The SONG is composed of plural SGNTs each of which is independently generated by shuffling the order of the given training data, and the output of the SONG is voted all outputs of the SGNTs. We allocate each of SGNTs to each of processors in the MIMD parallel computer. Experimental results show that the more the number of processors increases, the more the classification accuracy increases for all problems.
A wide variety of systems from many different fields use geospatial data to represent physical environments. These systems, and the data on which they operate, are not only becoming more complex, but they are increasi...
详细信息
A wide variety of systems from many different fields use geospatial data to represent physical environments. These systems, and the data on which they operate, are not only becoming more complex, but they are increasingly being connected together. Although each system may be interested in very different aspects of the environment, the geospatial data they use must be consistent and correlated with each other. As these "systems of systems" grow and as the amount of geospatial data gathered each day increases, the collection and preparation of these data has become the most time consuming and costly processes in the use of many of these systems. The solution to this problem lies in the creation of a geospatial dataset generation system that can efficiently collect, manipulate, and correlate geospatial data to the requirements of the end user. In this paper, we use the unique requirements of building geospatial datasets to derive a distributed software architecture for such a system.
We analyze the parallel processing in clusters of computers of a prediction method based on the improvement of Radial Basis Function (RBF) neural networks using matrix decomposition techniques such as the Singular Val...
详细信息
We analyze the parallel processing in clusters of computers of a prediction method based on the improvement of Radial Basis Function (RBF) neural networks using matrix decomposition techniques such as the Singular Value Decomposition (SVD) and the QR-cp factorization. parallel processing is required because of the extensive computation found in those techniques, but the reward is obtained in form of better prediction performance and less network complexity. This general prediction procedure (in the sequential version) was published in the technical literature previously, with a high degree of experimental success. parallelism is a convenient way to make this prediction module available for inexpensive operation within decision-making contexts. We discuss two alternatives of concurrency: parallel implementation of the prediction procedure over the ScaLAPACK suite, and the formulation of another parallel routine customized to a higher degree for better performance.
A cluster system can potentially provide features of high availability, extended scalability and fault tolerance. However, there are few cluster systems that have transactional capability with distributed lock manager...
详细信息
A cluster system can potentially provide features of high availability, extended scalability and fault tolerance. However, there are few cluster systems that have transactional capability with distributed lock managers to provide a single system image for transactional applications. We propose a distributed lock manager architecture called SSDLM for a cluster system, which is based on self-management of shared locks in a node to decrease lock management overhead in a cluster system. This paper describes the algorithm and the design of the lock management in SSDLM using data structures and processes. Finally, we test it through our experiments in terms of transaction response time and lock response time. The experimental results show that the overall performance improves with a small increase of nodes but becomes saturated soon, and that the locality of shared locks in SSDLM is effective for all experiments, and especially outstanding in the lock response time for read intensive transactions.
作者：
Xhafa, FatosDept. LSI
Univ. Politecnica de Catalunya Jordi Girona 1-3 08034 Barcelona Spain
In this paper we present an implementation of a generic memetic algorithm (MA) for the edge bi-connectivity augmentation problem (E2AUG). This problem is known for its applications to communication network design and ...
详细信息
In this paper we present an implementation of a generic memetic algorithm (MA) for the edge bi-connectivity augmentation problem (E2AUG). This problem is known for its applications to communication network design and in VLSI floor planning. Our starting point is a standard template for MA. We use a generic programming approach to obtain a C++ implementation of the MA template and instantiate it for the E2AUG problem. The implementation separates issues related to the memetic algorithm from those related to the E2AUG problem making thus possible (he reuse of the implementation for other problems. Moreover, the generic approach enabled us to easily implement the algorithm in parallel setting. We use known benchmarks in the literature for the problem to experimentally evaluate the implementation.
暂无评论