Transactional failure recovery for a distributed key-value store

With the advent of cloud computing, many applications have embraced the ensuing paradigm shift towards modern distributed key-value data stores, like HBase, in order to benefit from the elastic scalability on offer. However, many applications still hesitate to make the leap from the traditional relational database model simply because they cannot compromise on the standard transactional guarantees of atomicity, isolation, and durability. To get the best of both worlds, one option is to integrate an independent transaction management component with a distributed key-value store. In this paper, we discuss the implications of this approach for durability. In particular, if the transaction manager provides durability (e.g., through logging), then we can relax durability constraints in the key-value store. However, if a component fails (e.g., a client or a key-value server), then we need a coordinated recovery procedure to ensure that commits are persisted correctly. In our research, we integrate an independent transaction manager with HBase. Our main contribution is a failure recovery middleware for the integrated system, which tracks the progress of each commit as it is flushed down by the client and persisted within HBase, so that we can recover reliably from failures. During recovery, commits that were interrupted by the failure are replayed from the transaction management log. Importantly, the recovery process does not interrupt transaction processing on the available servers. Using a benchmark, we evaluate the impact of component failure, and subsequent recovery, on application performance.

Modelo de computación conexionista inspirado en las redes de procesadores evolutivos y su aprendizaje

La informática teórica es una disciplina básica ya que la mayoría de los avances en informática se sustentan en un sólido resultado de esa materia. En los últimos a~nos debido tanto al incremento de la potencia de los ordenadores, como a la cercanía del límite físico en la miniaturización de los componentes electrónicos, resurge el interés por modelos formales de computación alternativos a la arquitectura clásica de von Neumann. Muchos de estos modelos se inspiran en la forma en la que la naturaleza resuelve eficientemente problemas muy complejos. La mayoría son computacionalmente completos e intrínsecamente paralelos. Por este motivo se les está llegando a considerar como nuevos paradigmas de computación (computación natural). Se dispone, por tanto, de un abanico de arquitecturas abstractas tan potentes como los computadores convencionales y, a veces, más eficientes: alguna de ellas mejora el rendimiento, al menos temporal, de problemas NPcompletos proporcionando costes no exponenciales. La representación formal de las redes de procesadores evolutivos requiere de construcciones, tanto independientes, como dependientes del contexto, dicho de otro modo, en general una representación formal completa de un NEP implica restricciones, tanto sintácticas, como semánticas, es decir, que muchas representaciones aparentemente (sintácticamente) correctas de casos particulares de estos dispositivos no tendrían sentido porque podrían no cumplir otras restricciones semánticas. La aplicación de evolución gramatical semántica a los NEPs pasa por la elección de un subconjunto de ellos entre los que buscar los que solucionen un problema concreto. En este trabajo se ha realizado un estudio sobre un modelo inspirado en la biología celular denominado redes de procesadores evolutivos [55, 53], esto es, redes cuyos nodos son procesadores muy simples capaces de realizar únicamente un tipo de mutación puntual (inserción, borrado o sustitución de un símbolo). Estos nodos están asociados con un filtro que está definido por alguna condición de contexto aleatorio o de pertenencia. Las redes están formadas a lo sumo de seis nodos y, teniendo los filtros definidos por una pertenencia a lenguajes regulares, son capaces de generar todos los lenguajes enumerables recursivos independientemente del grafo subyacente. Este resultado no es sorprendente ya que semejantes resultados han sido documentados en la literatura. Si se consideran redes con nodos y filtros definidos por contextos aleatorios {que parecen estar más cerca a las implementaciones biológicas{ entonces se pueden generar lenguajes más complejos como los lenguajes no independientes del contexto. Sin embargo, estos mecanismos tan simples son capaces de resolver problemas complejos en tiempo polinomial. Se ha presentado una solución lineal para un problema NP-completo, el problema de los 3-colores. Como primer aporte significativo se ha propuesto una nueva dinámica de las redes de procesadores evolutivos con un comportamiento no determinista y masivamente paralelo [55], y por tanto todo el trabajo de investigación en el área de la redes de procesadores se puede trasladar a las redes masivamente paralelas. Por ejemplo, las redes masivamente paralelas se pueden modificar de acuerdo a determinadas reglas para mover los filtros hacia las conexiones. Cada conexión se ve como un canal bidireccional de manera que los filtros de entrada y salida coinciden. A pesar de esto, estas redes son computacionalmente completas. Se pueden también implementar otro tipo de reglas para extender este modelo computacional. Se reemplazan las mutaciones puntuales asociadas a cada nodo por la operación de splicing. Este nuevo tipo de procesador se denomina procesador splicing. Este modelo computacional de Red de procesadores con splicing ANSP es semejante en cierto modo a los sistemas distribuidos en tubos de ensayo basados en splicing. Además, se ha definido un nuevo modelo [56] {Redes de procesadores evolutivos con filtros en las conexiones{ , en el cual los procesadores tan solo tienen reglas y los filtros se han trasladado a las conexiones. Dicho modelo es equivalente, bajo determinadas circunstancias, a las redes de procesadores evolutivos clásicas. Sin dichas restricciones el modelo propuesto es un superconjunto de los NEPs clásicos. La principal ventaja de mover los filtros a las conexiones radica en la simplicidad de la modelización. Otras aportaciones de este trabajo ha sido el dise~no de un simulador en Java [54, 52] para las redes de procesadores evolutivos propuestas en esta Tesis. Sobre el término "procesador evolutivo" empleado en esta Tesis, el proceso computacional descrito aquí no es exactamente un proceso evolutivo en el sentido Darwiniano. Pero las operaciones de reescritura que se han considerado pueden interpretarse como mutaciones y los procesos de filtrado se podrían ver como procesos de selección. Además, este trabajo no abarca la posible implementación biológica de estas redes, a pesar de ser de gran importancia. A lo largo de esta tesis se ha tomado como definición de la medida de complejidad para los ANSP, una que denotaremos como tama~no (considerando tama~no como el número de nodos del grafo subyacente). Se ha mostrado que cualquier lenguaje enumerable recursivo L puede ser aceptado por un ANSP en el cual el número de procesadores está linealmente acotado por la cardinalidad del alfabeto de la cinta de una máquina de Turing que reconoce dicho lenguaje L. Siguiendo el concepto de ANSP universales introducido por Manea [65], se ha demostrado que un ANSP con una estructura de grafo fija puede aceptar cualquier lenguaje enumerable recursivo. Un ANSP se puede considerar como un ente capaz de resolver problemas, además de tener otra propiedad relevante desde el punto de vista práctico: Se puede definir un ANSP universal como una subred, donde solo una cantidad limitada de parámetros es dependiente del lenguaje. La anterior característica se puede interpretar como un método para resolver cualquier problema NP en tiempo polinomial empleando un ANSP de tama~no constante, concretamente treinta y uno. Esto significa que la solución de cualquier problema NP es uniforme en el sentido de que la red, exceptuando la subred universal, se puede ver como un programa; adaptándolo a la instancia del problema a resolver, se escogerín los filtros y las reglas que no pertenecen a la subred universal. Un problema interesante desde nuestro punto de vista es el que hace referencia a como elegir el tama~no optimo de esta red.---ABSTRACT---This thesis deals with the recent research works in the area of Natural Computing {bio-inspired models{, more precisely Networks of Evolutionary Processors first developed by Victor Mitrana and they are based on P Systems whose father is Georghe Paun. In these models, they are a set of processors connected in an underlying undirected graph, such processors have an object multiset (strings) and a set of rules, named evolution rules, that transform objects inside processors[55, 53],. These objects can be sent/received using graph connections provided they accomplish constraints defined at input and output filters processors have. This symbolic model, non deterministic one (processors are not synchronized) and massive parallel one[55] (all rules can be applied in one computational step) has some important properties regarding solution of NP-problems in lineal time and of course, lineal resources. There are a great number of variants such as hybrid networks, splicing processors, etc. that provide the model a computational power equivalent to Turing machines. The origin of networks of evolutionary processors (NEP for short) is a basic architecture for parallel and distributed symbolic processing, related to the Connection Machine as well as the Logic Flow paradigm, which consists of several processors, each of them being placed in a node of a virtual complete graph, which are able to handle data associated with the respective node. All the nodes send simultaneously their data and the receiving nodes handle also simultaneously all the arriving messages, according to some strategies. In a series of papers one considers that each node may be viewed as a cell having genetic information encoded in DNA sequences which may evolve by local evolutionary events, that is point mutations. Each node is specialized just for one of these evolutionary operations. Furthermore, the data in each node is organized in the form of multisets of words (each word appears in an arbitrarily large number of copies), and all the copies are processed in parallel such that all the possible events that can take place do actually take place. Obviously, the computational process just described is not exactly an evolutionary process in the Darwinian sense. But the rewriting operations we have considered might be interpreted as mutations and the filtering process might be viewed as a selection process. Recombination is missing but it was asserted that evolutionary and functional relationships between genes can be captured by taking only local mutations into consideration. It is clear that filters associated with each node allow a strong control of the computation. Indeed, every node has an input and output filter; two nodes can exchange data if it passes the output filter of the sender and the input filter of the receiver. Moreover, if some data is sent out by some node and not able to enter any node, then it is lost. In this paper we simplify the ANSP model considered in by moving the filters from the nodes to the edges. Each edge is viewed as a two-way channel such that the input and output filters coincide. Clearly, the possibility of controlling the computation in such networks seems to be diminished. For instance, there is no possibility to loose data during the communication steps. In spite of this and of the fact that splicing is not a powerful operation (remember that splicing systems generates only regular languages) we prove here that these devices are computationally complete. As a consequence, we propose characterizations of two complexity classes, namely NP and PSPACE, in terms of accepting networks of restricted splicing processors with filtered connections. We proposed a uniform linear time solution to SAT based on ANSPFCs with linearly bounded resources. This solution should be understood correctly: we do not solve SAT in linear time and space. Since any word and auxiliary word appears in an arbitrarily large number of copies, one can generate in linear time, by parallelism and communication, an exponential number of words each of them having an exponential number of copies. However, this does not seem to be a major drawback since by PCR (Polymerase Chain Reaction) one can generate an exponential number of identical DNA molecules in a linear number of reactions. It is worth mentioning that the ANSPFC constructed above remains unchanged for any instance with the same number of variables. Therefore, the solution is uniform in the sense that the network, excepting the input and output nodes, may be viewed as a program according to the number of variables, we choose the filters, the splicing words and the rules, then we assign all possible values to the variables, and compute the formula.We proved that ANSP are computationally complete. Do the ANSPFC remain still computationally complete? If this is not the case, what other problems can be eficiently solved by these ANSPFCs? Moreover, the complexity class NP is exactly the class of all languages decided by ANSP in polynomial time. Can NP be characterized in a similar way with ANSPFCs?

A generic persistence model for CLP systems (and two useful implementations)

This paper describes a model of persistence in (C)LP languages and two different and practically very useful ways to implement this model in current systems. The fundamental idea is that persistence is a characteristic of certain dynamic predicates (Le., those which encapsulate state). The main effect of declaring a predicate persistent is that the dynamic changes made to such predicates persist from one execution to the next one. After proposing a syntax for declaring persistent predicates, a simple, file-based implementation of the concept is presented and some examples shown. An additional implementation is presented which stores persistent predicates in an external datábase. The abstraction of the concept of persistence from its implementation allows developing applications which can store their persistent predicates alternatively in files or databases with only a few simple changes to a declaration stating the location and modality used for persistent storage. The paper presents the model, the implementation approach in both the cases of using files and relational databases, a number of optimizations of the process (using information obtained from static global analysis and goal clustering), and performance results from an implementation of these ideas.

A generic persistence model for CLP systems

This paper describes a model of persistence in (C)LP languages and two different and practically very useful ways to implement this model in current systems. The fundamental idea is that persistence is a characteristic of certain dynamic predicates (i.e., those which encapsulate state). The main effect of declaring a predicate persistent is that the dynamic changes made to such predicates persist from one execution to the next one. After proposing a syntax for declaring persistent predicates, a simple, file-based implementation of the concept is presented and some examples shown. An additional implementation is presented which stores persistent predicates in an external database. The abstraction of the concept of persistence from its implementation allows developing applications which can store their persistent predicates alternatively in files or databases with only a few simple changes to a declaration stating the location and modality used for persistent storage. The paper presents the model, the implementation approach in both the cases of using files and relational databases, a number of optimizations of the process (using information obtained from static global analysis and goal clustering), and performance results from an implementation of these ideas.

An abstract machine based execution model for computer architecture design and efficient implementation of logic programs in parallel.

The term "Logic Programming" refers to a variety of computer languages and execution models which are based on the traditional concept of Symbolic Logic. The expressive power of these languages offers promise to be of great assistance in facing the programming challenges of present and future symbolic processing applications in Artificial Intelligence, Knowledge-based systems, and many other areas of computing. The sequential execution speed of logic programs has been greatly improved since the advent of the first interpreters. However, higher inference speeds are still required in order to meet the demands of applications such as those contemplated for next generation computer systems. The execution of logic programs in parallel is currently considered a promising strategy for attaining such inference speeds. Logic Programming in turn appears as a suitable programming paradigm for parallel architectures because of the many opportunities for parallel execution present in the implementation of logic programs. This dissertation presents an efficient parallel execution model for logic programs. The model is described from the source language level down to an "Abstract Machine" level suitable for direct implementation on existing parallel systems or for the design of special purpose parallel architectures. Few assumptions are made at the source language level and therefore the techniques developed and the general Abstract Machine design are applicable to a variety of logic (and also functional) languages. These techniques offer efficient solutions to several areas of parallel Logic Programming implementation previously considered problematic or a source of considerable overhead, such as the detection and handling of variable binding conflicts in AND-Parallelism, the specification of control and management of the execution tree, the treatment of distributed backtracking, and goal scheduling and memory management issues, etc. A parallel Abstract Machine design is offered, specifying data areas, operation, and a suitable instruction set. This design is based on extending to a parallel environment the techniques introduced by the Warren Abstract Machine, which have already made very fast and space efficient sequential systems a reality. Therefore, the model herein presented is capable of retaining sequential execution speed similar to that of high performance sequential systems, while extracting additional gains in speed by efficiently implementing parallel execution. These claims are supported by simulations of the Abstract Machine on sample programs.

A new class of symbolic abstract neural nets

Starting from the way the inter-cellular communication takes place by means of protein channels and also from the standard knowledge about neuron functioning, we propose a computing model called a tissue P system, which processes symbols in a multiset rewriting sense, in a net of cells similar to a neural net. Each cell has a finite state memory, processes multisets of symbol-impulses, and can send impulses (?excitations?) to the neighboring cells. Such cell nets are shown to be rather powerful: they can simulate a Turing machine even when using a small number of cells, each of them having a small number of states. Moreover, in the case when each cell works in the maximal manner and it can excite all the cells to which it can send impulses, then one can easily solve the Hamiltonian Path Problem in linear time. A new characterization of the Parikh images of ET0L languages are also obtained in this framework.

An approach to a model of the visual cortex: disorders of vision

One of the most challenging problems that must be solved by any theoretical model purporting to explain the competence of the human brain for relational tasks is the one related with the analysis and representation of the internal structure in an extended spatial layout of múltiple objects. In this way, some of the problems are related with specific aims as how can we extract and represent spatial relationships among objects, how can we represent the movement of a selected object and so on. The main objective of this paper is the study of some plausible brain structures that can provide answers in these problems. Moreover, in order to achieve a more concrete knowledge, our study will be focused on the response of the retinal layers for optical information processing and how this information can be processed in the first cortex layers. The model to be reported is just a first trial and some major additions are needed to complete the whole vision process.

Symbolic vs substantive sustainability reporting: examining the role of company-level characteristics

As sustainability reporting (SR) practices have being increasingly adopted by corporations over the last twenty years, most of the existing literature on SR has stressed the role of external determinants (such as institutional and stakeholder pressures) in explaining this uptake. However, given that recent evidence points to a broader range of motives and uses (both external and internal) of SR, we contend that its role within company-level activities deserves greater academic attention. In order to address this research gap, this paper seeks to provide a more integrated perspective of both institutional and efficiency explanations of SR dynamics, as well as to highlight the role of company-level characteristics in explaining its contribution to sustainability management practices. More specifically, we suggest that substantive SR implementation can be predicted by assessing the level of fit between the organization and the SR framework being adopted. Building on this idea, our theoretical model defines three forms of fit (technical, cultural and political) and identifies organizational characteristics associated to each of these fits. Finally, implications for academic research, businesses and policy-makers are derived.