25 years ago the general shortage resp. high cost of computer
power placed serious limitations to simulation investigations of
data networks. To overcome this deficiency two versions of a special
simulation computer were designed: the first one, called ZUSI,
was based the principle ``time sequencing simulation'' [A19, A20]
and the second one, called DESC, on the principle ``event sequencing
simulation'' [A1, A2]. Both these computers
were multiprocessor systems with separate prosessors for typical
simulation tasks like random number generation, list processing and
statistical evaluation, which might be performed simultaneously.
Besides using standard simulation languages like SIMULA a
simulation system SIMPAS based on Pascal has been developed by H.
Tuchel and others [A18], which has been successfully operated for
many years on a computer Siemens 4004/151 computer (and followers)
owned by the chair and serving, among others, the extensive
LAN-simulation studies reported in [A30]. Furtheron the advanced
system SIC (``SImulation in C'') has been designed in C++ in order
to meet the needs of multiprocessor simulation with parallel
operation [A16, A17].
Several bad experiences due to deficiencies of
common pseudo-random number generators led, among others, to the
development of a new table driven random generator, called PURAN,
whose random bit were derived from physical phenomena like
atomic decay or random noise and were stored on a CD-memory
[A13, A14, A29]. Extensive
statistical tests
proved the quasi-ideal properties of this table generator.
Conventional methods for evaluating simulation output data
represented a further weak point of common simulation systems. To
overcome this deficiency the so called local correlation
coefficient has been introduced, which can be
associated to the distribution function F(x) of a random
x-sequence as a useful correlation function [A6, A34]
and is well suited for being measured
by statistical evaluation routines [A38].
On this basis the LRE-algorithm (LRE: Limited Relative Error) has
been developed [A35] and used for many years
as a reliable evaluation tool for analyzing correlated
output data and for controling the simulation run length. A
simplified version of this algorithm for discrete random
sequences [A40] has been recently introduced as part of the
so called RESTART/LRE-algorithm [A39] for
solving the simulation problem with respect to very
rare loss events in ATM-networks.
As years have passed by, some of the above listed research projects have
lost their former relevance. The speed-up achievement by the special
multiprocessor simulation computer DESC, for instance, has been outdated
in the meantime by the enormously grown speed of single processor work
stations. Also, most abilities of the formula evaluation system SCAFE have
been taken over by new output- and plot-facilities of an
algebraic system like MAPLE.
Other research results of the past years will keep their relevance.
There is the result, for instance, that the queueing discipline SRPT is
distinguished not only by its optimal mean delay feature but also by
other quantitative advantages, and that it can be effectively applied
in packet data networks: these facts of queueuing theory will remain
relevant for future teletraffic engineering and standardization work.
Also it can be presumed that the new statistical evaluation methods
will find broader acceptance in stochastic simulation and might even
help to solve the long debated objectivity problem in the field of statistics.
random bit were derived from physical phenomena like
atomic decay or random noise and were stored on a CD-memory
[A13, A14, A29]. Extensive
statistical tests
proved the quasi-ideal properties of this table generator.
has been introduced, which can be
associated to the distribution function F(x) of a random
x-sequence as a useful correlation function [A6, A34]
and is well suited for being measured
by statistical evaluation routines [A38].