Volume 15, 1996, No. 4


Universal Systems with Operations Related to Splicing

R. Freund, F. Wachtler

Abstract. Following some of the most recently obtained results on the computational universality of specific variants of H systems (e.g. see [6, 9, 13] we introduce systems based on new operations that are closely related to the splicing operation, i.e. we consider the operations of cutting a string at a specific site into two pieces with marking them at the cut ends and of recombining two strings with specifically marked endings. Whereas in the splicing of two strings these strings are cut at specific sites and the cut pieces are recombined immediately in a crosswise way, in the CR (cutting/recombination) systems introduced in this paper cutting can happen independently from recombining the cut pieces. Nevertheless, we shall show how a Turing machine computing a partial recursive function can be simulated by an equivalent mCR system (i.e. a CR system working in the multiset style, where the numbers of copies of all available strings are counted) with a finite set of cutting/recombination rules as well as with a finite set of axioms; in that way, from a universal Turing machine we obtain a universal mCR system.

 

Dissemination of Information in Vertex-Disjoint Paths Mode

J. Hromkoviè, R. Klasing, E.A. Stöhr

Abstract. The communication modes (one-way and two-way mode) used for disseminating information among processors of interconnection networks via vertex-disjoint paths in the communication step are investigated. The complexity of communication algorithms is measured by the number of communication steps (rounds). Since optimal broadcast and accumulation algorithms for these modes can be achieved in a straightforward way for almost all interconnection networks used, the paper concentrates on the gossip problem. The main results are listed below:

1.    Optimal gossip algorithms for paths, complete graphs and flakes in both modes.

2.    For hypercubes, cube-connected cycles, butterfly networks, etc., gossip algorithms which are only about O (log2log2n) rounds slower than the optimal gossip algorithms on complete graphic are designed.

Note that the results achieved have also practical applications, because the vertex/disjoint paths mode can be implemented in several hardware realisations of computing networks.

 

PSEE: A Tool for Parallel Systems Learning

E. Luque, R. Suppi, J. Sorribes, E. Cesar, J. Falguera, M. Serrano

Abstract. Programs for parallel computers of distributed memory are difficult to write, understand, evaluate and debug. The design and performance evaluation of algorithms is much more complex than the conventional sequential one. The technical know/how necessary for the implementation of parallel systems is already available, but a critical problem is in the handling of complexity. In parallel distributed memory systems the performance is highly influenced by factors as interconnection scheme, granularity, computing/communication ratio, algorithm topology, parallel languages, and operating system policies. Interaction between these factors is not easy to understand and often unpredictable. This paper presents the PSEE (Parallel System Evaluation Environment), an interactive graphical environment, which permits to study the behaviour of parallel distributed memory systems, as well as several enhancements to increase the goodness of this tool. PSEE is an easy-to-use environment that enables parallel distributed memory systems programmers take decisions about the behaviour program and parallel computer in terms such as scalability, tuning and performance of the underlying parallel machine.

 

A Cost Model for the Estimation Query Execution Time in a Parallel Environment Supporting Pipeline

M. Spiliopoulou, M. Hatzopoulos, C. Vassilakis

Abstract. We propose a model for the estimation of query execution time in an environment supporting bushy and pipelined parallelism. We consider a parallel architecture of processes having private main memories, accessing a shared secondary storage and communicating to each other via a network. For this environment, we compute the cost of query operators when processed in isolation and when in pipeline mode. WE use those formulae to incrementally compute the cost of a query execution plan from its components. Our cost model can be incorporated to any optimizer for parallel query processing that considers parallel and pipelined execution of the query operators.

 

Analysis and Design of a Relational Database Management System and Implementation of its Nucleus

M.C. Fernández-Baizán, A. García, M.M. González, C. Pérez-Llera, R. Portaencasa, E. Santos

Abstract. This paper describes the objectives and characteristics of a new relational database management system (RDBMS), that is being developed in the Technical University of Madrid. The project includes a development of a user interface based on the utilization of a universal relation, following the guidelines presented by Ullman in [62]. For this we will implement a graphical interface of similar characteristics to XWINDOWS. Furthermore, the RDBMS will be composed of a nonprocedural query language, an integrated data dictionary which contains database scheme in third normal form, a standard physical storage structure for all databases, an access optimization system based on a new relational algebra (t - algebra, defined by Garcia in [34] and published in [32]) and an integrity constraint control system. On the other hand, this RDBMS will allow the posterior development of a knowledge database system and an object-oriented database system, considering results of researchers that are going to be carried out in both areas.


Go To Contents