F1technical.net

Ft5fTL wrote:It seems its working with your solution. The only different thing is mine uses the cfd support version.

That's very interesting. Does yours work with the multicore option?

right, so I ran occfd with bluecore last night and came back to it this morning with a parallelisation issue. unfortunately it still exists. however, the simulation I ran yesterday proved that even if parallelisation works, single core is faster on my machine

On the OF section of CFD-online there are also many people complaining about installation issues on Windows machines. As I do not have one I do not know too much about it.
But would it not be easier to install Linux on a virtual machine and run OF in there?

LVDH wrote:On the OF section of CFD-online there are also many people complaining about installation issues on Windows machines. As I do not have one I do not know too much about it.
But would it not be easier to install Linux on a virtual machine and run OF in there?

Maybe it would be easier for those who have Linux knowledge. For pure windows folk like me, it's easier to run it single core overnight

LVDH wrote:But would it not be easier to install Linux on a virtual machine and run OF in there?

Yes, I used to do it, but I am not an expert with OpenFOAM and I can't extract the pressure differential when I run the simulation manually on Linux

Would it be possible to write it on a log?

Ft5fTL wrote:It seems its working with your solution. The only different thing is mine uses the cfd support version.

ha, I just checked the log of my simulation and it turns out that it used the cfd support version too.

in the occfd window, it detects bluecore, however, it uses cfd support if both are installed

Alonso Fan wrote:

Ft5fTL wrote:It seems its working with your solution. The only different thing is mine uses the cfd support version.

That's very interesting. Does yours work with the multicore option?

Multicore didnt work for me.

how may iterations does occfd do?

500 (preliminary run) + 800 (refined mesh run) = 1300 iterations

CAEdevice wrote:500 (preliminary run) + 800 (refined mesh run) = 1300 iterations

I'm running an analysis on my car and I'm 13 hours into a run, on iteration 593 simulation 1. How long do you think is left? Other family members also need to use my pc...

Alonso Fan wrote:

CAEdevice wrote:500 (preliminary run) + 800 (refined mesh run) = 1300 iterations

I'm running an analysis on my car and I'm 13 hours into a run, on iteration 593 simulation 1. How long do you think is left? Other family members also need to use my pc...

You can extrapolate how much time is left by looking at the time difference between 2 iterations in the .log file. Post-processing only takes a few mins compared to the simulation.

About to restart work on my car.

I am working away from home so I don't have access to my powerhouse (6 core Xeon 2.8ghz, 24gb ram) just my laptop (4 core i7 1.6ghz, 8gb ram). I can't leave it running all night as the fan screams when it gets hot.

Also I have a lot of ground to make up on those who have already got a car designed and developed.

I have made a list of various designs I want to test (5 different front fenders, 3 floor designs and 4 cooling solutions before I even begin to look at wings) so using AWS allows me to design one, send it off for analysis and then once they are done pick the best two to take forward.

I was wondering how long a run costs (time & money) on AWS for a similar length run on a desktop? Can you throw more money to get the time down (if 8hr run cost 25p does a 1hr run cost £2)?

Is there a correlation between higher and lower resolution meshes? So two meshes both have X facets. One is 5% better (downforce, drag, whatever) than the other. If I half the number of facets does the relationship still hold that one is 5% better? If so when does this relationship break down? X/4, X/8?

No you cant throw more money at AWS as Julien has it set up on a specific virtual machine (m3 xlarge) which I find takes around 9 to 10 hours. Would have to ask Julien if he could set it to run on the next level up and see how much time it cut off to see if it was cost effective. Costs about $3 a run (you get charge for the storage of the files too so its worth downloading them and deleting them off the server asap)

Thanks for letting me know about the issues with surfaceOrient. My mistake (again). This aspect should be fixed in patch 11, download from http://www.khamsinvirtualracecarchallen ... .1-011.zip

RicME85 wrote:No you cant throw more money at AWS as Julien has it set up on a specific virtual machine (m3 xlarge) which I find takes around 9 to 10 hours. Would have to ask Julien if he could set it to run on the next level up and see how much time it cut off to see if it was cost effective. Costs about $3 a run (you get charge for the storage of the files too so its worth downloading them and deleting them off the server asap)

I choose the virtual machine to try to "maximise" cost-effectiveness. The first simulation (run 0) is run on a single core even if the VM has multiple core - as it is a small mesh it may be detrimental to run on multiple core. The meshing (snappyHexMesh) of the run_1 is done in sequential. When I tried it I could not get a benefit from running this in parallel. The meshing typically takes around 1 hour and drives the memory requirement (snappyHexMesh is memory hungry).

I have a plan to migrate to the more recent instance family but did not get time to do this.

One of the benefits of AWS is that you can send a few simulations at the same time - while if you are running on your local machine you might be typically be limited to one or two run at once.