|
Dear Rémi and TRNSYS users,
We recently did some experimenting with a dual-core
processor. Our aim was actually to find out if the hard disk writing of a
computer could be saturated by large TRNSYS outputs and thus become a
bottle-neck in simulations with big output files (and it seems like not easily
with standard modern computers).
But...
To benefit from a dual core processor a program
must be multi-tasking, i.e. it must create multiple threads for the operating
system (OS) which the OS can then run in parallel. TRNSYS does not create
multiple threads, so if you are running only one simulation on a dual core
processor, it will run just as fast as if you would run it on a single core.
However, you can run two TRNSYS simulations
simultaneously, and the OS will then run each simulation on its own
core. This means that you can run e.g. two parametric studies (or
divide one into two) at the same time and thus halve the needed time.
BUT TAKE CARE: you must run two decks from
separate directories and see that also the output files end up in different
directories!!! (or change the names) Otherwise the simulations will write
to same files, the output data of the first started will be replaced by the
later started.
I have no knowledge of how much you can benefit of
dual core processors when calling other programs from TRNSYS or with plugins.
But based on what is stated above, I'd say that you will benefit only, if the
other programs can do something useful while TRNSYS is solving. If TRNSYS
solver pauses for feedback from the other program and vice versa for each
iteration or time-step, you will probably have no notable perfomance boost. But,
for example, if the other program does data processing or writing to the
hard disk each time step after it has given feedback to TRNSYS then you
might benefit, as this will be done by the other core and TRNSYS can
continue a bit earlier. You have to try and see...
Also, I don't know what happens with communication
between TRNSYS and the other programs if there are several instances running...
I have only tried with TRNSYS without external connections.
I'd also be very interested to hear if other people
have more experience on this. What comes to RAM, you should check how much you
are using with current simulations. You can do that with the windows performance
monitor by checking how much RAM is used and how much it changes when you start
a simulation. Additional RAM will only help if the OS is writing to the paging
file (virtual memory on hard disk) while you are running the simulation.
1GB for the computer should be more than enough, but it depends of course on
what other programs and system processes are running simultaneously...
TRNSYS does not eat memory as e.g. FEM/CFD does.
What comes to processor speeds, the sim time should
be somewhat directly proportional to processor speed, so faster is better. That
is, if there are no other bottle-necks on the motherboard (bus
speeds, cache etc.). Usually in good quality commercial computers (or built by
experts) there should be no problems with processor not being optimized with
motherboard. (but changing a processor to your existing motherboard is another
thing...)
Hope that helps at least as a starting
point...
Cheers
-----------------------------------------------------------------------------
Janne Paavilainen MSc, PhD student Researcher in Energy and Environmental engineering ----------------------------------------------------------------------------- Solar Energy Research Center SERC Dalarna University College SE-781 88 Borlänge Sweden ----------------------------------------------------------------------------- tel +46 (0)23 778728 fax +46 (0)23 778701 e-mail: jip@du.se ----------------------------------------------------------------------------- www.du.se www.serc.se www.eses.org www.uni-kassel.de/fb15/ite/solar/solnet/ -----------------------------------------------------------------------------
|