|
[Sponsors] |
![]() |
![]() |
#1 |
Member
Song Young Ik
Join Date: Apr 2022
Location: South Korea
Posts: 59
Rep Power: 4 ![]() |
I am planning to use Azure cyclecloud and openpbs to create HPC Openfoam server.
What I am curious about is, how do they work inside cluster. In single node cfd server, MPI will take care of cores and distribute work. In cluster, each seperate nodes will have their own OS. Can MPI send order between different nodes? Does OpenFOAM have capability to do such calculation traveling across nodes? What does pbs actually do between nodes? Do I need to access every compute node and install OpenFOAM every single one of them, or it isn't? If latter is correct, I guess I should set server node as accelerated network supported VM so... I have literally no idea how cluster works |
|
![]() |
![]() |
![]() |
![]() |
#2 |
Senior Member
Lucky
Join Date: Apr 2011
Location: Orlando, FL USA
Posts: 5,761
Rep Power: 66 ![]() ![]() ![]() |
The whole point of clouds is to not have to worry about such things...
pbs is just a script used to submit jobs to the job scheduler, which will then assign tasks to the individual cpus. In your example pbs will tell the job scheduler to queue up x instances of your favorite foam solver. It goes to the dungeon master who then plays a little waiting game. Once it has decided the outcome of the game it will then send your foam solver tasks to each cpu. In a single node all these tasks get assigned to the same node. In a cluster, some get sent to other nodes. If the other nodes are assigned to accept and run these tasks then they will do so. If not, reject them. Generally you have one shared disk where all your junk is installed and all the nodes access this same disk and run the executables from the same directory. It is extremely uncommon to have an installation on each node because it is a headache that you don't want to experience dealing with asymmetric configurations, i.e. potentially different openfoam versions, potentially different compilations and builds used on each machine, and they will almost certainly say F U when you try to run them. Single node machines like your desktop also have a job scheduler, you just never have to deal with it because you don't usually need to manage your own computer. |
|
![]() |
![]() |
![]() |
![]() |
#3 |
New Member
beckham
Join Date: Oct 2024
Posts: 6
Rep Power: 2 ![]() |
OpenPBS acts as a job scheduler, managing resource allocation and job execution across the cluster. You typically do not need to install OpenFOAM on every compute node individually; instead, you can set up a Doodle Jump shared file system or use containerization to streamline the deployment.
|
|
![]() |
![]() |
![]() |
Tags |
azure, cloud, cluster, mpi |
|
|
![]() |
||||
Thread | Thread Starter | Forum | Replies | Last Post |
Applicability of a specific software in CFD modelling | ivan-s | Main CFD Forum | 0 | April 22, 2014 14:30 |
Software Language for Coding CFD | vsookrah | Main CFD Forum | 5 | September 15, 2013 20:34 |
What is the optimal (economically and user friendly) CFD software for my application? | Rame | Main CFD Forum | 0 | August 10, 2009 06:30 |
Student looking for CFD software | Chris | Main CFD Forum | 6 | January 21, 2005 04:20 |
Where do we go from here? CFD in 2001 | John C. Chien | Main CFD Forum | 36 | January 24, 2001 22:10 |