Optimizing communication by compression for Multi-GPU Scalable Breadth-First Searches

Debido al creciente aumento del tamaño de los datos en muchos de los actuales sistemas de información, muchos de los algoritmos de recorrido de estas estructuras pierden rendimento para realizar búsquedas en estos. Debido a que la representacion de estos datos en muchos casos se realiza mediante estructuras nodo-vertice (Grafos), en el año 2009 se creó el reto Graph500. Con anterioridad, otros retos como Top500 servían para medir el rendimiento en base a la capacidad de cálculo de los sistemas, mediante tests LINPACK. En caso de Graph500 la medicion se realiza mediante la ejecución de un algoritmo de recorrido en anchura de grafos (BFS en inglés) aplicada a Grafos. El algoritmo BFS es uno de los pilares de otros muchos algoritmos utilizados en grafos como SSSP, shortest path o Betweeness centrality. Una mejora en este ayudaría a la mejora de los otros que lo utilizan. Analisis del Problema El algoritmos BFS utilizado en los sistemas de computación de alto rendimiento (HPC en ingles) es usualmente una version para sistemas distribuidos del algoritmo secuencial original. En esta versión distribuida se inicia la ejecución realizando un particionado del grafo y posteriormente cada uno de los procesadores distribuidos computará una parte y distribuirá sus resultados a los demás sistemas. Debido a que la diferencia de velocidad entre el procesamiento en cada uno de estos nodos y la transfencia de datos por la red de interconexión es muy alta (estando en desventaja la red de interconexion) han sido bastantes las aproximaciones tomadas para reducir la perdida de rendimiento al realizar transferencias. Respecto al particionado inicial del grafo, el enfoque tradicional (llamado 1D-partitioned graph en ingles) consiste en asignar a cada nodo unos vertices fijos que él procesará. Para disminuir el tráfico de datos se propuso otro particionado (2D) en el cual la distribución se haciá en base a las aristas del grafo, en vez de a los vertices. Este particionado reducía el trafico en la red en una proporcion O(NxM) a O(log(N)). Si bien han habido otros enfoques para reducir la transferecnia como: reordemaniento inicial de los vertices para añadir localidad en los nodos, o particionados dinámicos, el enfoque que se va a proponer en este trabajo va a consistir en aplicar técnicas recientes de compression de grandes sistemas de datos como Bases de datos de alto volume o motores de búsqueda en internet para comprimir los datos de las transferencias entre nodos.---ABSTRACT---The Breadth First Search (BFS) algorithm is the foundation and building block of many higher graph-based operations such as spanning trees, shortest paths and betweenness centrality. The importance of this algorithm increases each day due to it is a key requirement for many data structures which are becoming popular nowadays. These data structures turn out to be internally graph structures. When the BFS algorithm is parallelized and the data is distributed into several processors, some research shows a performance limitation introduced by the interconnection network [31]. Hence, improvements on the area of communications may benefit the global performance in this key algorithm. In this work it is presented an alternative compression mechanism. It differs with current existing methods in that it is aware of characteristics of the data which may benefit the compression. Apart from this, we will perform a other test to see how this algorithm (in a dis- tributed scenario) benefits from traditional instruction-based optimizations. Last, we will review the current supercomputing techniques and the related work being done in the area.

Constructs and evaluation strategies for intelligent speculative parallelism - armageddon revisited

This report addresses speculative parallelism (the assignment of spare processing resources to tasks which are not known to be strictly required for the successful completion of a computation) at the user and application level. At this level, the execution of a program is seen as a (dynamic) tree —a graph, in general. A solution for a problem is a traversal of this graph from the initial state to a node known to be the answer. Speculative parallelism then represents the assignment of resources to múltiple branches of this graph even if they are not positively known to be on the path to a solution. In highly non-deterministic programs the branching factor can be very high and a naive assignment will very soon use up all the resources. This report presents work assignment strategies other than the usual depth-first and breadth-first. Instead, best-first strategies are used. Since their definition is application-dependent, the application language contains primitives that allow the user (or application programmer) to a) indícate when intelligent OR-parallelism should be used; b) provide the functions that define "best," and c) indícate when to use them. An abstract architecture enables those primitives to perform the search in a "speculative" way, using several processors, synchronizing them, killing the siblings of the path leading to the answer, etc. The user is freed from worrying about these interactions. Several search strategies are proposed and their implementation issues are addressed. "Armageddon," a global pruning method, is introduced, together with both a software and a hardware implementation for it. The concepts exposed are applicable to áreas of Artificial Intelligence such as extensive expert systems, planning, game playing, and in general to large search problems. The proposed strategies, although showing promise, have not been evaluated by simulation or experimentation.

The ciao prolog system

Ciao is a public domain, next generation multi-paradigm programming environment with a unique set of features: Ciao offers a complete Prolog system, supporting ISO-Prolog, but its novel modular design allows both restricting and extending the language. As a result, it allows working with fully declarative subsets of Prolog and also to extend these subsets (or ISO-Prolog) both syntactically and semantically. Most importantly, these restrictions and extensions can be activated separately on each program module so that several extensions can coexist in the same application for different modules. Ciao also supports (through such extensions) programming with functions, higher-order (with predicate abstractions), constraints, and objects, as well as feature terms (records), persistence, several control rules (breadth-first search, iterative deepening, ...), concurrency (threads/engines), a good base for distributed execution (agents), and parallel execution. Libraries also support WWW programming, sockets, external interfaces (C, Java, TclTk, relational databases, etc.), etc. Ciao offers support for programming in the large with a robust module/object system, module-based separate/incremental compilation (automatically -no need for makefiles), an assertion language for declaring (optional) program properties (including types and modes, but also determinacy, non-failure, cost, etc.), automatic static inference and static/dynamic checking of such assertions, etc. Ciao also offers support for programming in the small producing small executables (including only those builtins used by the program) and support for writing scripts in Prolog. The Ciao programming environment includes a classical top-level and a rich emacs interface with an embeddable source-level debugger and a number of execution visualization tools. The Ciao compiler (which can be run outside the top level shell) generates several forms of architecture-independent and stand-alone executables, which run with speed, efficiency and executable size which are very competive with other commercial and academic Prolog/CLP systems. Library modules can be compiled into compact bytecode or C source files, and linked statically, dynamically, or autoloaded. The novel modular design of Ciao enables, in addition to modular program development, effective global program analysis and static debugging and optimization via source to source program transformation. These tasks are performed by the Ciao preprocessor ( ciaopp, distributed separately). The Ciao programming environment also includes lpdoc, an automatic documentation generator for LP/CLP programs. It processes Prolog files adorned with (Ciao) assertions and machine-readable comments and generates manuals in many formats including postscript, pdf, texinfo, info, HTML, man, etc. , as well as on-line help, ascii README files, entries for indices of manuals (info, WWW, ...), and maintains WWW distribution sites.