id: 00830377
dt: b
an: 00830377
au: Hockney, Roger W.
ti: The science of computer benchmarking.
so: Software - Environments - Tools, 2. Philadelphia, PA: SIAM, Society for
    Industrial and Applied Mathematics. xiv, 129 p. \$ 21.25 (1995).
py: 1995
pu: Philadelphia, PA: SIAM
la: EN
cc: 
ut: computer benchmarking; parallel computers performance tests; DUSD method
ci: 
li: doi:10.1137/1.9780898719666
ab: There is not too much research on computer benchmarking, and especially
    parallel computers performance tests. This is surprizing as the
    performance/cost radio should guide application and system design. It
    is indeed difficult to evaluate the computer performances taking into
    account the gap between hardware possibilities and software
    abstractions. Another source of performance differencerates for the
    same computer consists in the very different type and complexity of
    applications, from image processing and text retrieval to PDE solvers.
    Roger Hockney devoted much of his time and efforts to developing a
    model for computer performance, the original $(r,n {1 \over 2})$ model,
    and several benchmarks included now within the PARKBENCH set. The
    author argues against improper used performance metrics like speedup
    and efficiency. Moreover, he introduces the DUSD method which can help
    in understanding how the performance of a program or of a benchmark
    varies with the number of processors and with the problem size. Another
    interesting point is how to extrapolate results from one computer to
    another one with different hardware characteristics. The book is full
    of experimental results, traced on several architectures and sustaining
    the author’s model and methodologies. This 100 pages book is
    worthstudying by postgraduate students and systems engineers.
rv: D.Grigoras (Iaşi)