|Posted on Thursday, November 30, 2000 - 6:26 am: |
We have a powerful SGI server with plenty
of disk space and memory for Gaussian jobs.
I installed WebMo on a linux-server where
we run the www-applications and interfaces.
While NOT reading so carefully I did not
understand that Gaussian has to be installed
on the same server. Are there any plans to
enable sending jobs from linux WebMo to
Gaussian on an SGI?
|Posted on Friday, December 08, 2000 - 1:52 am: |
It should be relatively easy. First, create a script to run g98 on SGI (it should define $g98root first, just like your user .login does). Then copy input file from Linux to SGI (it is easier when SGI HD is remotely mounted on Linux). Finally, run that script remotely "[user@linux]% rsh sgi yourscript inputfile -l user" from linux, instead of SGI local command "[user@sgi]% g98 inputfile".
|William Polik (Polik)
|Posted on Friday, December 08, 2000 - 1:24 pm: |
WebMO is designed to run on the same machine as the computational chemistry program to which it interfaces (Gaussian, MOPAC, GAMESS). Thus, the most straightforward solution is to install a webserver onto your SGI.
We are considering mechanisms for having WebMO interface to multiple computational servers, but this is architecturally challenging and certainly won't happen any time soon, if ever.
However, Artem makes an interesting suggestion. If you set up file sharing correctly (so the SGI computational server could see the Linux web server's filesystem), perhaps WebMO could invoke Gaussian on the SGI using a rsh call (the environmental variables would have to be set up correctly). We have not tested this configuration in any way. If you were interested and adventurous, the starting point would be WebMO's gaussian.cgi script, which invokes gaussian.
|Posted on Monday, March 05, 2001 - 2:10 pm: |
The obvious thing to try is to run a queue system on the machine acting as the web server. The Gaussian job would be submitted to the queue rather than being run interactively. The NQS system, for example, is free and manages a network queue system. One could set the queue to include four machines, for example. The nqs queue manager would pass the task to whichever machine had spare capacity. I do not have intimate knowledge of the design of WebMO, so it may be a challenge to track the progress of each task and retrieve the results. I suspect one could work around this issue, though.
I am still contemplating the use of WebMO in our program, but if we do use WebMO we will certainly want the possibility of running computations on one of any number of machines. (We also have the parallel version of Gaussian for a Beowulf cluster, and it would seem easy to implement this with WebMO.)
|Posted on Tuesday, March 06, 2001 - 10:30 am: |
Development for a separate web-server/computational-server architecture is underway for a future release of WebMO. This architecture will also allow the WebMO sever to support multiple computational clients, as requested. For example, one would either pick a server to submit the job to, or simply the first available server. Most likely, we will avoid the use of a "standard" queuing system for compatibility reasons.