5.2 Large Systems Studies Using Classical Dynamics
173
one has λ ≡ 8 · 10
8 m. In supersonic and hypersonic flows rarefaction is characterized
by the Tsein’s parameter, that is equivalent to the product of the Knudsen and the
Mach (M) number (KnM or M
2 /Re) with Re being the Reynolds number.
The DSMC method models the flow of a fluid in terms of colliding bodies made
by a large number of molecules and solving the related Boltzmann equation. Bodies
are moved through a simulation of physical space in a realistic manner that is directly
coupled to physical time. Interbody and body-surface collisions can be calculated
using probabilistic, phenomenological and collisional models. The method finds
application to several technologies including the estimation of the Space shuttle
reentry aerodynamics and the modeling of microelectronic-mechanical systems using
appropriate interfaces (see Refs. [121–124]).
5.3 Supercomputing and Distributed Computing
Infrastructures
5.3.1 High-Performance Versus High-Throughput
Computing
In general, computational chemistry has grown by exploiting at any time the best
performing compute machines available. As already mentioned, the real rise of computational chemistry to the dignity of separate discipline, has occurred only with
the advent of the so-called mainframes (one (CPU) to many (users) compute platforms) based on ad hoc designed advanced CPUs and characterized in the field of
molecular sciences for the ability of carrying out high-performance off-line FORTRAN number crunching dominant calculations. This has occurred in the second
part of the 20th century when the mainframes were offering increasing compute
power on single CPU architectures by continuously improving circuitry efficiency
(shorter clock-period, faster electronics, larger buses and higher miniaturization),
components quality (communication bandwidth, size and speed of caches, size and
efficiency of memories), optimized use of the processor (multiprogramming, time
sharing, look ahead and prefetching) often driven by progress in science research.
As just mentioned, the reference language was FORTRAN that evolved through
different versions to the highly popular FORTRAN IV (1961), FORTRAN 66, and
FORTRAN 77. More recent versions are those of 1990, 1995, 2003, 2008, and 2015
increasingly tailored to suit the HPC needs.
Further speed-enhancing progress was associated with architectural changes to the
organization of different levels of cache memories and with the design of innovative
languages and operating systems. The most significant advances were associated with
concurrent execution at both operation and instruction level, multiple functional units
and pipelines, super- and multiscalar microarchitectures, long instruction words, etc.
At the same time, the exploitation of architectural innovation at such micro level
significantly enhanced the possibility of faster parallel runs of highly coupled pro-
173
one has λ ≡ 8 · 10
8 m. In supersonic and hypersonic flows rarefaction is characterized
by the Tsein’s parameter, that is equivalent to the product of the Knudsen and the
Mach (M) number (KnM or M
2 /Re) with Re being the Reynolds number.
The DSMC method models the flow of a fluid in terms of colliding bodies made
by a large number of molecules and solving the related Boltzmann equation. Bodies
are moved through a simulation of physical space in a realistic manner that is directly
coupled to physical time. Interbody and body-surface collisions can be calculated
using probabilistic, phenomenological and collisional models. The method finds
application to several technologies including the estimation of the Space shuttle
reentry aerodynamics and the modeling of microelectronic-mechanical systems using
appropriate interfaces (see Refs. [121–124]).
5.3 Supercomputing and Distributed Computing
Infrastructures
5.3.1 High-Performance Versus High-Throughput
Computing
In general, computational chemistry has grown by exploiting at any time the best
performing compute machines available. As already mentioned, the real rise of computational chemistry to the dignity of separate discipline, has occurred only with
the advent of the so-called mainframes (one (CPU) to many (users) compute platforms) based on ad hoc designed advanced CPUs and characterized in the field of
molecular sciences for the ability of carrying out high-performance off-line FORTRAN number crunching dominant calculations. This has occurred in the second
part of the 20th century when the mainframes were offering increasing compute
power on single CPU architectures by continuously improving circuitry efficiency
(shorter clock-period, faster electronics, larger buses and higher miniaturization),
components quality (communication bandwidth, size and speed of caches, size and
efficiency of memories), optimized use of the processor (multiprogramming, time
sharing, look ahead and prefetching) often driven by progress in science research.
As just mentioned, the reference language was FORTRAN that evolved through
different versions to the highly popular FORTRAN IV (1961), FORTRAN 66, and
FORTRAN 77. More recent versions are those of 1990, 1995, 2003, 2008, and 2015
increasingly tailored to suit the HPC needs.
Further speed-enhancing progress was associated with architectural changes to the
organization of different levels of cache memories and with the design of innovative
languages and operating systems. The most significant advances were associated with
concurrent execution at both operation and instruction level, multiple functional units
and pipelines, super- and multiscalar microarchitectures, long instruction words, etc.
At the same time, the exploitation of architectural innovation at such micro level
significantly enhanced the possibility of faster parallel runs of highly coupled pro-