Switches for FePEST?

Hi again,

On bootup, I'd like all my machines on my network to be slaves and listen on a specific port so I don't have to set up each one manually each time. I'd like to also specify this port before each machine is in service (currently stuck on default port on FePEST startup). I think this is possible with the PEST Utilities pslave.exe but I'd like to stick with the FePEST GUI interface (or FePESTc.exe ;) ).

Thanks again,

Adam

Hi Adam,

Good to know that you are intensively testing new features in FePEST! Here some ideas you could use in your workflow:

- FePEST can be started in server mode directly from the command line:
e.g. [color=red]"C:\Program Files\WASY\FEFLOW 6.2\bin64\fepest62.exe" -mode=server[/color]

- You can alter the port on which the FePEST server (where the slaves run)  using a registry setting. This should be created in the following manner:
e.g. [color=red][HKEY_CURRENT_USER\Software\DHI-WASY\FEFLOW Parameter Estimation] --> "Port"=dword:000013bb[/color]

- This registry setting must also be set on the client (master) computer. Please note, the input is in hexadecimal but you can also choose decimal using regedit. Deal with careful any modification in the Registry Editor, incorrect changes can affect FePEST installation. The port on which BeoPEST runs, is set in the FePEST Settings dialog on the Parallelization page.

- PSLAVE.EXE is part of the parallel pest solution PPEST. FePEST uses BEOPEST (third-party software) and, as stated above, this port is specified in the FePEST Settings.

Hope I could clarify your doubts. Let me know you have other questions under specific circumstances of your network.

Cheers,

Carlos

Ah, everything is clear now. Thank you so much! Parallelization is definitely helping me figure out what I'm doing wrong with FePEST much faster :)

P.S. I'd also like to change the timeout to a lower value, as it currently is way too high and I need to forcibly close or stop the application in case something goes wrong...

Adam

Hi Adam,

Here we go...

The default Timeout is 30000 msecs (= 30 seconds), which is the time the client or server waits until assuming that the requested operation failed, and repeats it. This default value is also the default of the underlying library function used by FePEST. It may be possible to lower this without side effects, but beware, if the network is responding badly, or if the FEM is very large then a low timeout could mean client/server communication can never succeed. You can modify its value with a registry entry named Timeout.

In general you can track errors using the log FePEST file: %TEMP%\.fepest62\log.txt

To forcibly stop PEST, you should use red stop button. Sometimes you may have to wait a number of seconds until it works. This is because is carrying out the last FEFLOW run. The black Kill button is more brutal, and kills the process and so should be used only if the red stop button fails.

Cheers,

Carlos

Yeah, I'm having issues with the beopest process on my slaves, where it continues to run for an indefinite time... I was thinking the 30000 msecs was actually seconds, which is why I asked. I'm having problems because one particular slave wasn't connected to the license server (which I have since fixed, but have been able to reproduce the problem).

I'm fairly certain that it does not stop and continues to search for the license server beyond 30 seconds and repeats indefinitely or does something similar. When I try to restart the slave, I get a socket error because beopest is still running in the background. However, when I kill everything using the black button on the host then restart the slave, the slave still continues where it left off. I think it works again if once I close FePEST.

I've had a similar problem with the host too, where it tries to finish what its doing even if the slave explodes. If I press the stop or pause button and reload the .fps file, it is still stuck indefinitely. I'm thinking there's a problem with the timeout, but maybe I'm just trying too hard to break it :)

Adam

Hi Adam,

If a client command does not complete FePEST gives up after <Timeout> seconds, but then repeats the failed command. The idea behind this is to try to make FePEST tolerant of temporary network problems.

We have tried starting a FePEST server with an invalid FEFLOW license and ran a parallelized FePEST problem. The server log repeated the message “not licensed”, just as you mentioned. However, pressing stop on the client cancelled the PEST run, after a short delay. (In some cases it may be necessary to stop the server first). So, we weren't able to reproduce your problem, just by making the license unavailable. I guess what your workflow must have been slightly different.

Cheers,

Carlos

Thanks Carlos. In my case, I have a unsaturated flow model with 2M elements and 30 parameters, so that might be why the processes still remain after I've killed it and closed FePEST on both host and slave.

Edit: Actually, I think that they "appear" to be running but they aren't. I'm not sure, but it's working so I'm not going to worry about it.

Thanks again,

Adam