sharing users id over nodes on a cluster

potac · February 6, 2023, 14:33

Although my question is not related to a hardware, I would like to know if somebody can give me a hint related to sharing user ids over a linux cluster.
I have managed to connect several pcs to a cluster. As s job manager I use SLURM. As I was installing the single nodes, there is my user account on each of them. Everything works surprisingly well and I can submit a job and run a calculation (OpenFOAM so far). I would like to open the cluster to other users by creating new user account on the master node. However, when any new user submits a job, the job crashes. From the error message I can understand that openmpi would like to have users ids over all the nodes. If I create the new user over all the nodes manually all works.
I wonder whether I have to share user ids over all the nodes for instance by using 398 DS with ldap or there is another approach to solve this.
Thx.

wkernkamp · February 6, 2023, 16:54

For mpi to work over a cluster, you have to be able to log in with ssh without password. So any method you find that allows that from one of your nodes to another, will also enable the mpi runs when implemented across the cluster.

potac · February 20, 2023, 12:18

Do you know what is the most common way it is done on clusters with many users?
I am checking ssh with Host-based Authentication now if that's the way to go. Thx.

https://en.wikibooks.org/wiki/OpenSS...Authentication

February 6, 2023, 14:33	sharing users id over nodes on a cluster	#1
potac Member Jan Potac Join Date: Aug 2009 Posts: 30 Rep Power: 16	Although my question is not related to a hardware, I would like to know if somebody can give me a hint related to sharing user ids over a linux cluster. I have managed to connect several pcs to a cluster. As s job manager I use SLURM. As I was installing the single nodes, there is my user account on each of them. Everything works surprisingly well and I can submit a job and run a calculation (OpenFOAM so far). I would like to open the cluster to other users by creating new user account on the master node. However, when any new user submits a job, the job crashes. From the error message I can understand that openmpi would like to have users ids over all the nodes. If I create the new user over all the nodes manually all works. I wonder whether I have to share user ids over all the nodes for instance by using 398 DS with ldap or there is another approach to solve this. Thx.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
SIMPLE algorithm	SamR	Main CFD Forum	19	April 26, 2019 13:06
Problems running on multiple nodes in a cluster	naveen k s	OpenFOAM Programming & Development	3	February 14, 2019 17:36
Cluster ID's not contiguous in compute-nodes domain. ???	Shogan	FLUENT	1	May 28, 2014 15:03
F_NNODE returns double number of nodes	falkon	Fluent UDF and Scheme Programming	0	March 8, 2014 06:58
CFX4.3 -build analysis form	Chie Min	CFX	5	July 12, 2001 23:19

February 6, 2023, 16:54		#2
wkernkamp Senior Member Will Kernkamp Join Date: Jun 2014 Posts: 321 Rep Power: 12	For mpi to work over a cluster, you have to be able to log in with ssh without password. So any method you find that allows that from one of your nodes to another, will also enable the mpi runs when implemented across the cluster.

February 20, 2023, 12:18		#3
potac Member Jan Potac Join Date: Aug 2009 Posts: 30 Rep Power: 16	Do you know what is the most common way it is done on clusters with many users? I am checking ssh with Host-based Authentication now if that's the way to go. Thx. https://en.wikibooks.org/wiki/OpenSS...Authentication