Jordi Bataller

Flexible Management of Consistency and Availability of Networked Data Replications

We describe a family of three replication protocols, each of which can operate in three different modes of consistency. The protocols are tailored to satisfy the availability demands of interconnected databases that have a high degree of data locality. The protocols accomplish a grade of transaction completion which does not compromise availability, and ensure the consistency of replicas also if a transaction needs to be aborted. Flexibility of query answering is understood as optimizing the tradeoff between consistency and availability, i.e., between correctness and timeliness of query answering. This is achieved by choosing an appropriate protocol alternative, and changing the consistency mode of operation during the session, as appropriate for a given transaction.

A Parallel Method for Impulsive Image Noise Removal on Hybrid CPU/GPU Systems

A parallel algorithm for image noise removal is proposed. The algorithm is based on peer group concept and uses a fuzzy metric. An optimization study on the use of the CUDA platform to remove impulsive noise using this algorithm is presented. Moreover, an implementation of the algorithm on multi-core platforms using OpenMP is presented. Performance is evaluated in terms of execution time and a comparison of the implementation parallelised in multi-core, GPUs and the combination of both is conducted. A performance analysis with large images is conducted in order to identify the amount of pixels to allocate in the CPU and GPU. The observed time shows that both devices must have work to do, leaving the most to the GPU. Results show that parallel implementations of denoising filters on GPUs and multi-cores are very advisable, and they open the door to use such algorithms for real-time processing.

A Fuzzy Metric in GPUs: Fast and Efficient Method for the Impulsive Image Noise Removal

The implementation of image correction algorithms on the CUDA platform is a relatively new field. Although the platform is easy to program, it is not easy to optimize the applications due to the number of decisions that have to be made. This paper reports an optimization study on the use of the CUDA platform to remove impulsive noise in images using fuzzy metric and the concept of peer group. The texture memory is used to speed up the access to data. In order to get the maximum bandwidth on the GPU memory, a strategy based on storing each pixel in 4 bytes is proposed.

Enhancing the availability of networked database services by replication and consistency maintenance

We describe a middleware platform for maintaining the consistency of replicated data called COPla (Common Object Platform). The purpose of replication is to enhance the availability of data and services in distributed database networks. That is orthogonal to recovery strategies of backed-up snapshots, logs and other measures to alleviate database downtimes. A range of different consistency modes ensures the correctness of replicated data.

Replication for Web-based collaboration

We suggest to increase the dependability of Web-based collaboration systems by the distributed replication of underlying data. We describe the architecture of the middleware package DIRECS. It supports the consistency of replicated data and thereby increases responsiveness, availability and failure resilience of collaborative systems.

Adaptable Distributed Shared Memory: A Formal Definition

In this paper we introduce a new model of consistency for distributed shared memory called Mume. It offers only the essentials to be considered as a shared memory system. This allows an efficient implementation on a message passing system, and due to this, it can be used to emulate other memory models.

Constructive and adaptable distributed shared memory

Distributed shared memory (DSM) is a paradigm for programming distributed systems, which provides an alternative to the message passing model. DSM offers the agents of the system a shared address space through which they can communicate with each other. The main problem of a DSM implementation on top of a message passing system is performance. Performance of an implementation is closely related to the consistency the DSM system offers: strong consistency (all agents agree about how memory events happen) and is more expensive to implement than weak consistency (disagreements are allowed). There have been many DSM systems proposals, each one supporting different consistency levels. Experience has shown that no one is well suited for the whole range of problems. In some cases, strong consistent primitives are not needed, while in other cases, the weak semantics provided are useless. This is also true for different implementations of the same memory model, since performance is also affected by the data access patterns of the applications. We introduce a novel DSM model called Mume. Mume is a low level layer close to the level of the message passing interface. The Mume interface provides only the minimum requirements to be considered as a shared memory system. The interface includes three types of synchronization primitives, namely total ordering, causal ordering and mutual exclusion. This allows efficient implementations of different memory access semantics, accommodating particular data access patterns.

Peer group and fuzzy metric to remove noise in images using heterogeneous computing

In this paper, we report a study on the parallelization of an algorithm for removing impulsive noise in images. The algorithm is based on the concept of peer group and fuzzy metric. We have developed implementations using Open Multi-Processing (OpenMP) and Compute Unified Device Architecture (CUDA) for Graphics Processing Unit (GPU). Many sequential algorithms have been proposed to remove noise, but their computational cost is excessive for real-time processing of large images. We developed implementations for a multi-core CPU, for a multi-GPU (several GPUs) and for a combination of both. These implementations were compared also with different sizes of the image in order to find out the settings with the best performance. A study is made using the shared memory and texture memory to minimize access time to data in GPU global memory. The result shows that when the image is distributed in multi-core and multi-GPU a greater number of Mpixels/second are processed.