Buenas Daniel, If you don't need to run interactive jobs (srun, salloc) there should not be any issue. You only need the client packages and the config files on the submit hosts. The submit hosts must be able to reach the slurmctld host, but they do not need to see the internal cluster.
Regards, Carlos On Wed, Jun 14, 2017 at 12:57 PM, Daniel Ruiz Molina < daniel.r...@caos.uab.es> wrote: > > Hi Miguel, > > No, I don't need to connect to SLURM clusters, so I have only one > slurmctld. What I need is connect some computers from "1.1.1.0" network to > a HPC Clusters in "1.1.2.0" network. From computers I only need submit jobs > that will be executed inside clusters. So my question (or doubt) is what I > need to share between two networks for allowing submit jobs from one side > to the other side. > > > El 12/06/2017 a las 7:47, Miguel Gila escribió: > >> Hola Daniel, >> >> My replies inline below. >> >> >> On 9 Jun 2017, at 18:21, Daniel Ruiz Molina <daniel.r...@caos.uab.es> >>> wrote: >>> >>> >>> Hello, >>> >>> I have this scenario and I need to know if it would be possible for >>> SLURM to work. >>> >>> I have some computers that will act as "submit" SLURM hosts; in other >>> words, in that hosts won't run batch or interactive jobs. That computers >>> are configured with a network as 1.1.1.0/24 for example. >>> >> Do I understand here submit host = login host? >> >> On the other side of my scenario, I manage a HPC Cluster configured in a >>> network like 1.1.2.0/24. I can connect using SSH from submits computers >>> to HPC Cluster server with no problem >>> >>> I need to connect my submits hosts with my entire HPC Cluster for >>> allowing people submit jobs from hosts and, then, execute that jobs in the >>> cluster. From server, I will share "/slurm" folder via NFS for all the >>> compute nodes inside the cluster, but also I could share "/slurm" with >>> submit hosts (opening their ports in iptables). I suppose I need to >>> configure in "slurmd.conf" all my nodes: HPC compute nodes and submit hosts. >>> >> As far as I know, the only thing you need to physically share between the >> two slurmctlds is the StateSaveLocation if you want failover. Other than >> that you need to: >> - make sure they all use the same Munge key, if you’re using Munge. >> - make sure all systems are reachable in the configured Slurm ports. >> - make sure all systems have the same defined users. >> >> There is no concept of ‘routing node' as in other scheduling systems, >> submission hosts need to be able to reach compute nodes and control >> daemons. If this cannot be an option, it is possible to have ssh wrappers >> here and there… but it becomes ugly very fast. >> >> Will run SLURM if it is configured for both networks? >>> >> As said, yes, if you have all-to-all routing enabled on the configured >> ports. >> >> Could anybody answer me? >>> >>> Thanks. >>> >> Hope this helps. >> >> Cheers, >> Miguel >> >> -- -- Carles Fenoy
