Page 296 - The Combined Finite-Discrete Element Method

LARGE SCALE COMBINED FINITE-DISCRETE ELEMENT SIMULATIONS 279

Say that job A takes one day running on its own, and job B takes one day running on its
own. Thus if only one machine is available, after day one the results of the job A will
be ready for post-processing, and after day two the results of job B will be available for
post-processing. Should the jobs be run simultaneously, the best case scenario is that after
two days the results of both jobs are available. Very often, the situation is much worse.
Jobs A and B, when run simultaneously, compete not only for CPU time but also for
RAM space. If both jobs involve small scale combined finite-discrete element simulations,
it is most likely that they will both fit nicely within the available RAM space, and no
paging will be necessary. However, if each job requires more than 50% of the available
RAM space, both of them will not be able to fit within the available RAM at the same
time, and some data from the in-core database will have to sit on the hard disk. This will
bring a familiar situation of the frequent need to access the hard disk. The result will be
a significant, sometimes tenfold, increase in run time.
It is evident that the RAM requirements of combined finite-discrete element simulations
must be taken very seriously. There are several things that can be done:

• Avoid multitasking, i.e. running more than one combined finite-discrete element simu-
lation on the same machine at the same time.
• Run only combined finite-discrete element simulations that require less space than the
available RAM space on the given machine.
• Design the in-core database in such a way that the size of the database is reduced to a
minimum. Two approaches are available, namely an object orientated in-core database
and a relational in-core database.

9.1.2 Minimising CPU requirements

Minimisation of CPU requirements can be achieved by:
• more efficient algorithms,
• more efficient implementations, and
• the use of faster hardware including parallel, distributed and grid computing options.

9.1.3 Minimising storage requirements

Combined finite-discrete element simulations require a massive number of results to be
stored during the simulation. Hard disk requirements can easily run into hundreds and
thousands of gigabytes. These can reduce the CPU efficiency, but can also make certain
combined finite-discrete element simulations impossible without a large disk storage space
being readily available. The solution is data compression before any results are stored
onto the hard disk.

9.1.4 Minimising risk

Large scale combined finite-discrete element simulations can run for days, even weeks.
Restart procedures are necessary to reduce the risk of loosing valuable CPU times due