Hi,
Very new to BCM. I have a SLURM cluster and I want to integrate it with BCM. But I am running into several issues (or maybe limitations/inflexibilities in BCM) during the installations. Here they are:
- Cluster uses two kinds of node base names. One is for compute nodes and other for user sessions. BCM only allows defining a single node base name.
- Cluster has servers from several hardware manufacturers but BCM requires you to specify a single manufacturer.
- It is unclear the options to choose in the “compute nodes” section during the installation process for a cluster that is already live and no new nodes need to be added. Shall I choose the number of existing nodes or choose 0 since no new nodes need to be provisioned at the time of BCM installation.
- In the “BMC Configuration” section, since the cluster has servers from more than one manufacturer, choosing 1 BMC network type isn’t possible. But installation requires you to choose 1.
- Since there’s already a DHCP server that exists in the cluster, it is unclear how I can configure BCM DHCP server to avoid DHCP conflicts or configure it to use existing networks, subnets and VLANs instead of creating new ones.
Would really appreciate if someone can provide some clarity on this or point to the appropriate documentation.
Cheers,
Hi there! Welcome to BCM and we hope you will have a good experience with it notwithstanding these set backs. Let me try to answer these as best I can:
node base names
You can always change the names. When you initially set up the cluster it will say things like node001-node004 as the prefix. But you can always change those afterwards. So you can have node001-node999 and k8s-node-01-k8s-node-02. Just during the initial setup in the iso image it seems to only allow one name. When I set it up I usually overwrite all of that after the initial head node installation.
multi-vendor/ manufacturer
This is again the initial wizard. After you are installed you can edit/create/modify the nodes. BCM does support multi manufacturers.
no new nodes to be added
I’m curious about this one because isn’t the idea that BCM will be managing the compute nodes? If they are already provisioned do you intend to have BCM reprovision them? You can always select 0 or 1 (if it doesn’t allow you to select less than 1) and then remove the configuration later after its installed.
BCM Configuration 1 BCM Network
Yeah the wizard tries to simplify things and is usually good for 80-90% of the use cases but as mentioned above, you can just install it with some defaults and then go in and modify everything to make it work with your environment afterwards.
DHCP
This is another curiosity I have of your use case. Typically you want BCM to manage your network because then when a node PXE boots the BCM DHCP server will serve it the right files to tftpboot from. If you are doing something different then you can always turn off DHCP and only serve static IP addressese by disabling unknown DHCP serving. In cmsh:
cmsh
network
use internalnet
set lockdowndhcpd yes
commit
Hi @vbenincosa,
Thanks very much for your detailed response and clarifying the behaviour for node base names, multi-vendor and BCM configuration.
On the subject of no new nodes , we have a functional cluster without a cluster manager. So having a centralized provisioning and management tool is nice to have. Furthermore, there is always need to add more nodes to the cluster and even upgrade the cluster node OS to a different major OS (something that I suspect can be significant work without a cluster manager). Also, I was under the impression that BCM can manage pre-deployed servers (via agent service etc.) which appears to not be the case. I suspect I will have to gradually redeploy OS on servers, this time with BMC, for the cluster to be onboarded.
On DHCP, yes I wanted to use DHCP but finetune it so it works in conjunction with the existing DHCP server. Since, from your response, it looks like that is possible to do post-installation.
Thanks again, I will proceed with the installation following your helpful response.