Re: [Nanodesigner-developers] Design Question

[<sw...@us...>]

Fuji Hakayito wrote:

> <>I've been reading up on JavaSpace technology and it's really
> interesting, I look forward to seeing how it works with the project.
> Anyways, I have a question about the Nanodesigner architecture: Why do
> we need two separate clients for Designing (DC) and Computing (CC)? I
> understand from the design description that the CC will only be
> responsible for processing job requests (workunits). Is the intention
> here to utilize machines dedicated for processing and processing only?
> Or could we have one machine running both the CC and the DC clients
> concurrently?
> Let me know if you understand my question and what you think.

Using two or more systems is only necessary in case there is a need for=20
large amounts of raw processing power not uncommon in molecular=20
modelling and design. Obviously to take advantage of parallel processing=20
you also need algorithms that can be parallelized which is a problem on=20
its own. So in case parallelizing is not possible or not desirable the=20
processing needs to be done on the Design Client itself.
I think it should be possible to use both on the same machine although I=20
am not sure what the benefit would be in this case, except maybe for=20
debugging purposes.

You are correct, it is the intention to use machines for processing only.
Mind you that over time the newly designed molecules will become larger=20
and larger so obviously the processing power needed will become more=20
extensive. To give an idea about it: an ab initio (read: exact)=20
molecular modelling calculation of a molecule of 10 atoms or so takes=20
all the processing power of the biggest supercomputers. There is no need=20
to be alarmed however because I doubt it will ever be needed for the=20
goals we have set ourselves. There are a lot of other algorithms, less=20
accurate of course, but who will do quite nicely for the purposes of=20
molecular nanotechnology and other fields. Those kind of algorithms only=20
give approximations, true, but they do it a lot faster than ab initio=20
algorithms.

More details about such an environment can be found on=20
http://math.nist.gov/mcsd/savg/parallel/screen/index.html
Javier was kind enough to point me to it earlier on. I have contacted=20
the person in charge, William George, and he replied that their software=20
will be open source and available around the end of this year. I see no=20
reason why we couldn=92t adapt it to our goal. Also it would be=20
interesting to adapt it I think. No need to reinvent stuff. That holds=20
true for other parts of the software also. We only need to be careful=20
about license issues.

Does this answer your questions Fuji?

>Also, let me know if you guys have any GUI chat sessions.  I
>unfortunately don't have the Yahoo messenger, but I do have the
>following:
>ICQ: 18059890
>MSN: atm...@ho...
>(I like IRC as well)
> =20
>
There has been no chat session for the UI group yet. You are in the=20
UTC-7 time zone if I am not mistaken, Javier is in the UTC time zone and=20
Chandresh is in the UTC+4 time zone I believe. It might be difficult to=20
arrange a chat session and I must confess I didn=92t take into account th=
e=20
time difference while creating the groups. I only took into account your=20
apparent interests.

Val=E8re