Igor Tartalja

Classifying software-based cache coherence solutions

The authors propose a classification for software solutions to cache coherence in shared memory multiprocessors and show how it can be applied to more completely understand existing approaches and explore possible alternatives.

EM/sup 3/: a taxonomy of heterogeneous computing systems

The field of heterogeneous computing is growing rapidly. New concepts and systems appear daily. Hence, it is important to fit each new contribution into its proper place in the puzzle called heterogeneous computing. This is possible only if an adequate taxonomy/classification exists, one that can show whether or not a new system is heterogeneous, and if so, what kind of heterogeneity it exhibits. We propose a new taxonomy that shows the relative position of each and every heterogeneous system in the overall computer systems world. The proposed taxonomy is intended to be both broad enough to encompass all existing heterogeneous systems and simple enough to be easily accepted. Consequently, our taxonomy includes only four classes of computer systems. We propose that computer systems be classified as follows: SESM (single execution mode, single machine model); SEMM (single execution mode, multiple machine models), MESM (multiple execution modes, single machine model), and MEMM (multiple execution modes, multiple machine models).

An approach to dynamic software cache consistency maintenance based on conditional invalidation

Introduces a class of software protocols for the maintenance of cache consistency in multiprocessors with shared main memory and private caches. These protocols are designed to be built into the operating system primitives for mutual exclusion. The approach is based on a dynamic decision about invalidation of the shared segment copy residing in the private cache, at the moment of entering into the appropriate critical region. The authors gradually introduce three consistency schemes.<<ETX>>

HFPaC: GPU friendly height field parallel compression

In this paper, we present a novel method for fast lossy or lossless compression and decompression of regular height fields. The method is suitable for SIMD parallel implementation and thus inherently suitable for modern GPU architectures. Lossy compression is achieved by approximating the height field with a set of quadratic Bezier surfaces. In addition, lossless compression is achieved by superimposing the residuals over the lossy approximation. We validated the method’s efficiency through a CUDA implementation of compression and decompression algorithms. The method allows independent decompression of individual data points, as well as progressive decompression. Even in the case of lossy decompression, the decompressed surface is inherently seamless. In comparison with the GPU-oriented state-of-the-art method, the proposed method, combined with a widely available lossless compression method (such as DEFLATE), achieves comparable compression ratios. The method’s efficiency slightly outperforms the state-of-the-art method for very high workloads and considerably for lower workloads.

Domino tiling: a new method of real-time conforming mesh construction for rendering changeable height fields

In this paper we present a novel GPU-oriented method of creating an inherently continuous triangular mesh for tile-based rendering of regular height fields. The method is based on tiling data-independent semi-regular meshes of non-uniform structure, a technique that is quite different from other mesh tiling approaches. A complete, memory efficient set of mesh patterns is created by an off-line procedure and stored into the graphics adapters memory at runtime. At rendering time, for each tile, one of the precomputed mesh patterns is selected for rendering. The selected mesh pattern fits the required level of details of the tile and ensures seamless connection with other adjacent mesh patterns, like in a game of dominoes. The scalability potential of the proposed method is demonstrated through quadtree hierarchical grouping of tiles. The efficiency is verified by experimental results on height fields for terrain representation, where the method achieves high frame rates and sustained triangle throughput on high resolution viewports with sub-pixel error tolerance. Frame rate sensitivity to real-time modifications of the height field is measured, and it is shown that the method is very tolerant and consequently well tailored for applications dealing with rapidly changeable phenomena represented by height fields.

test: tools for evaluation of students' tests-a development experience

The purpose of the project that we describe in this paper is to provide teachers with efficient means of generating and scoring tests with multiple choice answers, inevitable in evaluating students knowledge at massive examinations. The idea is to maintain a database containing classified test questions, together with precisely defined criteria for selecting questions for an exam. A particular test can be composed automatically, or semiautomatically with active participation of the teacher. The final test document is produced in multiple versions, which represent various permutations of questions and answers. For each version, an answer key is also produced, which is used by an automatic scoring tool, that supports arbitrary hand-held or flat-bed scanner, for scanning answer sheets. This paper describes a set of software tools that we developed, in order to provide teachers with the efficient assistance in this process, on the commonly available PCs running Windows.

Scowl: a tool for characterization of parallel workload and its use on SPLASH-2 application suite

Concentrates on the problem of defining and measuring parameters that characterize typical behavior of parallel applications targeted to distributed shared memory (DSM) systems and shared-memory multiprocessors (SMPs). These parameters can be used as input to various models for performance evaluation in this research area. Furthermore, typical application behaviors can be recognized, which can help to generate new ideas for improvements to memory consistency protocols, adapting them to specific application characteristics. Our study encompasses a variety of parameters, such as frequencies of operations of various access types (private read/writes, shared read/writes, lock operations, barrier operations), the average number of accessed blocks per interval, the average number of modified words, etc. The results presented in this paper are based on the SPLASH-2 (Stanford Parallel Applications for SHared Memory) application suite. The developed instrumentation tool Scowl, along with the applied simulation environment Limes, are publicly available and applicable for performing measurements on other parallel applications as well.

Memory consistency models for shared memory multiprocessors and DSM systems

The use of systems with multiple processors that support shared memory programming paradigm is rapidly increasing nowadays. Possible buffering, pipelining, and optimization of shared memory accesses, as well as the existence of multiple copies of shared variables in these systems, may cause specific implications that can not be understood just as an intuitive extension of an uniprocessor memory model. Therefore, the memory consistency model formally specifies the memory system behavior to be expected by the programmer. This paper reveals the essence of several memory consistency models: sequential, processor, weak, release (with eager and lazy implementation), and entry. It also provides definitions and a set of examples that underline differences between particular models. Results of several performance evaluation studies are also discussed.

A classification of eLearning tools based on the applied multimedia

This paper proposes a classification that should help understanding key aspects of multimedia application in eLearning tools. The classification tries to cover important aspects of multimedia application in eLearning tools: communication channels and exchange of different types of contents throughout the channels, understanding in communication, and the ways of object manipulation in the user-tool interaction. Types of contents are classified according to the senses they affect on both sides of communication channels. The paper presents characteristics of 30 representative tools through a uniformly structured text. The presented tools are analyzed according to the proposed classification. A view of the future challenges, based on an analysis of the global trends in the area, is given.

An approach to characterization of parallel applications for DSM systems

The paper concentrates on the problem of defining and measuring parameters that characterize behavior of parallel applications targeted to DSM (distributed shared memory) systems. Results are based on the SPLASH-2 application suite. The developed characterization tool Scopa, along with applied simulation environment Limes, are publicly available, and appropriate for performing measurements on other parallel applications, as well.