Suggestion for Early Stage Deployment

Community Support
1

Hi,

Newbie here, I would like to test out Open Nebula on a lab test. I have 3 servers that are going to be used as the test.

  • First server running as the front-end with the following specs,
    Quad-core Xeon, 8 GB RAM, 240 GB SSD, 2 Gigabit NIC

  • Second and the third server specs,
    8 Cores Xeon, 128 GB RAM, 2 TB HDD, 2 Gigabit NIC

I would like to setup the server to be HA on the 2 compute nodes, and also running Ceph. As for testing purposes, we only have 1 Gigabit connectivity for each nodes. Can it be run? We also want to test whether the public IP will be able to reach the instances, and also creating private networks for our clients later on.

The production purpose is to serve IaaS for our clients.

Thank you,

Regards,

2

hello, you should have at least 3nodes in HA cluster. You can self-host frontend, so no need for extra node

3

Hi,

Thank you very much for your response!

I will plan to deploy one more node for the cluster, but can it be deployed later on? After the current setup is production ready.

Btw, by self-hosted front end you mean running it on one of the compute kvm nodes?

Regards

4

Hello, you should research more about ceph, but what I read and heard, 3 nodes are absolute recommended minimum.

About self-hosted frontend - I run frontend as VM using KVM, but manage it by pacemaker. So it start after cluster is up. I have plan to setup three VMs and run frontend in HA using new RAFT algorithm. Each VM on different node.

You save physical server and also power consuption…

5

I can confirm that the opennebula including the frontend runs fine in VMs. I have a 3 node HA cluster setup (RAFT) running like that. My master nodes run on vmware and manage a bunch of KVM hosts but the management instances can obviously run on the kvm nodes as well.

My ceph knowhow is as well limited as we are still working on getting it up and running but ceph needs some nodes and to run real workload on it you will want to use physical nodes that hold your osds. Vms for the monitors and so on are ok and for testing you can run the osds in vms too but if you really want to run serious load on ceph you will need hardware for it.

On the networking side you should be fine. I have some smaller node connected with only 1G and those host 5+ instances. It highly depends on you workload though. If the instances are very active you can run into issues fast. The instances produce network traffic and the datastore will do that as well so if your workload does not involve a lot of these tasks you will be fine but heavy IO instances will soon saturate your 1G uplink.

As my instance networks get their ips via dhcp and not opennebula i can’t offer any advice on your accessibility question. But basically if you can route the traffic to your instances you will be fine.