A method for describing computational science.
The concepts and terminology of high performance distributed scientific computing are blurred. In (Tanenbaum & Van Renesse, 1985) it was mentioned that "not everyone agrees on what they mean by the term distributed system", in spite that "everyone agrees that distributed systems are going to be very important in the future". Today the problem has been extrapolated by new concepts like grid scientific computing and home-grown interconnected TFLOPS systems.
In (Sameh et. al., 1996), CSE is described as being driven by scientific and engineering applications in the apex of a pyramid with benchmarking, computer architecture, algorithms and system software in the base square of enabling technologies. The author considers this structure not suited for solving the disagreements on terminology nor for identifying old solutions to new rising problems.
Several taxonomies have been proposed to clarify the various types of computer systems and architectures among which (Flynn, 1972), (Tanenbaum & Renesse, 1985) and (Dasgupta, 1990) are more representative. In (Anton & Cretu, 2009) the authors applied artificial intelligence techniques on ~6000 articles linked to scientific computing in order to understand how they relate to each other; three major clusters were discovered within the data set.
In this paper several taxonomies of computer architectures are compared and their most fundamental approaches are identified. Based on the previous research, an original structure for describing computational science and engineering is proposed.
The new structure separates the information into "concept vs platform" layers with cross-like interconnected fields, and finally merges them together in a compass. The structure can be used to organize the scientific papers within the computational science databases making it easier to spot old solutions to new problems, and can be used to solve the terminological barriers inherited by the "high performance distributed scientific computing" domains, mostly from within marketing decisions.
The research enables future investigations in the problem of describing computational science and engineering. Future work should address a formal description of the structure in three-dimensional space, based on oriented graphs, suitable for browsing a database of scientific information from various angles.
2. TAXONOMICAL APPROACH
In (Capello, 2007) the major forces that shaped the evolution of application-specific processors were identified and observed to have remained unchanged. The author agrees on this viewpoint and acknowledges that most of the barriers overcome in massively parallel processors (MPPs) are identical with those that currently shape the evolution of computational clusters.
Table 1 presents the fundamental approaches to computational system taxonomies. The table contains the references that either started a new taxonomical approach or are the most representative for their kind. There are other approaches on computational system architectures, most of them derived from Table 1 or too similar in order to create a new branch.
Diverse definitions of parallel architectures have been proposed (Duncan, 1990). Still, the terms sometime remain blurred even when the context is clear and opinions may vary. In this paper the author will try to pinpoint fundamental angles that best describe computational science and engineering.
3. PROPOSED STRUCTURE
Based on (Anton & Cretu, 2009), Figure 1 depicts the three major clusters identified in scientific computing related articles. It is easy to notice that the physical resolver represents the platform of computing and that machine speech together with the discretized model can be merged into a conceptual section. When the concept is applied to a platform, a solution is computed.
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
In Figure 2, the platform is described by the dark circles and the concept by the larger objects. The platform circles are smaller in order to depict the spatial superposition of the concept layer on top. It is detailed according to fundamental computer science enabling subfields. The computer architecture describing the physical resolver shapes the programming language & paradigm. The machine is abstracted by the operating system and enabled by network topologies. Network topologies can be internal or external, for MPPs or computer clusters.
The conceptual level starts by repeteadly observing the physical phenomenon and eventually shaping the mathematical model that describes it. Through the theory of computation the model is transformed into a discretized algorithm with abstract data types; this also constitutes the first level of implementation in the form of "machine speech" (see Figure 1 or PL&P in Figure 2) and is to be understood as formal language by the platform.
Once the correct circle is identified, it is easy to resolve the disagreements on terminology. For example: distribution, grid and clusters are issues of network topology, from the platform layer. Distribution is a concept linked to fault tolerance, transparency and liveness (ie. memory).The grid term is thus borrowed from the electrical energy supply networks and it refers to the topology of the distribution, being a higher-level network for supplying resources like information and services, in a fault-tolerant and redundant manner. The term also encapsulates concepts of authentication, audit, authorization and access.
Clusters refer to a group of (mostly) homogenous networked computers which are "aware" of each other (ie. opaque) and which use message passing to mimic massively parallel processors for intense scientific computations.
Parallelism and concurrency are issues of programming language and paradigms, and supercomputing is defined by computer architecture. Parallelism is a set of actions and events that do not take place in sequence, but simultaneously.
Concurrency is a higher-level concept that is linked to PL&P and provides encapsulated mechanisms for portions of programs to be executed in parallel.
The MPPs and their more cost-effective cluster counterparts represent the de facto platforms for scientific computing--see (Feitelson, 2005). Supercomputing is defined as scientific computing on computer clusters or MPPs, and thus a supercomputer is to be understood as any of the two scientific computing platforms.
In (Kung, 1982) old computer design issues that fit into the network topology section present solutions to modern cluster design problems.
This paper introduces a new structure for describing computational science and engineering, which can be used to organize the information within scientific databases related to the forementioned domain, in order to quickly spot old available solutions to more recent problems and resolve the disagreements on terminology.
The relationships depicted with arrows are yet to be formally defined, and future work should address a graph oriented structure in three-dimensional space, suited for navigating through a scientific database from various angles.
In order to formally describe the attachment of information to one of the eight classes, a weighted algorithm needs to be developed.
Opensource folklore compares the publishing of scientific results with the delivery of software source code. Computational science and engineering, along with its yet to be improved navigational approaches should be addressed by an open science initiative, based on open technologies.
Anton, A.-A. & Cretu, V.-I. (2009), Unsupervised exploration of scientific articles, Proceedings of the 5th International Symposium on Applied Computational Intelligence and Informatics, pp.539-544, Timisoara, May 2009
Capello, P. (2008), Application-specific Processor Architecture: Then and Now, Journal of Signal Processing Systems, Vol.53, No.1-2, (November 2008), pp. 197-215, ISSN:1939-8018
Dasgupta, S. (1990), A Hierarchical Taxonomic System for Computer Architectures, Computer, Vol. 23, No. 3, (March 1990), pp. 64-67, ISSN:0018-9162
Duncan, R (1990), A survey of parallel computer architectures, Computer, Vol.23, No.2, (February 1990), pp.5-16
Feitelson, D. G. 2005, The Supercomputer Industry in Light of the Top500 Data, Computing in Science and Engineering, Vol.7, No.1, (Jan/Feb 2005), pp.42-47
Flynn, M. (1972), Some Computer Organizations and Their Effectiveness, IEEE Transactions on Computers, Vol.C-21, (September 1972), pp.948
Giloi, W. K. (1994), Parallel Supercomputer Architectures and their Programming Models, Parallel Computing, Vol.20, No.10-11, (November 1994), pp.1443-1470, ISSN:0167-8191
Kung, H. T. (1982), "Why Systolic Architectures?", Computer, Vol.15, No.1, (January 1982), pp.47-46,ISSN:0018-9162
Sameh, A.; Cybenko, G.; Kalos, M.; Neves, K.; Rice, J.; Sorensen, D. & Sullivan, F (1996), Computational science and engineering, ACM Computing Surveys, Vol.28, No.4, (December 1996), pp.810-817, ISSN:0360-0300
Tanenbaum, A. S. & Van Renesse, R. (1985), Distributed operating systems, ACM Computing Surveys, Vol. 17, No.4, (December 1985), pp. 419-470, ISSN:0360-0300
Weper, R.; Zehendner, E. & Erhard, W. (1999), p: Hierarchical modeling of parallel architectures, Proceedings of the 7th Euromicro Workshop on Parallel and Distributed Processing, pp.233-240, Funchal, February 1999
Tab. 1. Fundamental approaches on taxonomy Classification Modelling Simulation (Flynn, (Tanenbaum (Dasgupta (Giloi, (Weper et. 1972) & Van 1990) 1993) al, 1999) Renesse, 1985) Information Angle of Chemistry Hierarchy Object stream transparency description
|Printer friendly Cite/link Email Feedback|
|Publication:||Annals of DAAAM & Proceedings|
|Date:||Jan 1, 2009|
|Previous Article:||Influence of kinematic parameters on the deterministic vibrations of the viscoelastic linear connecting rod: part of a rod lug mechanism.|
|Next Article:||Six inextensible wire device for determination of position-orientation matrix of a robot's mobile reference system.|