Newbie questions about frontend and ssh datastore

Community Support
1

I’ve been reading the docs and planning to give opennebula a try, but I have some doubts I would like to solve before… so I need your expertise…

My plan is to configure a server as frontend with storage for the images datastore and add 4 to 6 hypervisor nodes (KVM).

The hypervisor nodes would have from 30 to 40 cores each; about 256/512 GB of RAM and would start with 4 TB of raid 10 SSD storage (maybe ZFS).

There would not be any shared datastore as I would configure Filesystem Datastore with ssh transfer mode.

That cluster would host about 100 virtual machines of different sizes and purposes

That is what I’ve thougth it would be a good configuration, but I’m shure there are many things I did not take into account. Could I have your thougts about this configuration? Is there something I’m doing wrong? would you configure it in another way?

I also would like to know if one Front-end will be enough, and how much CPU/RAM would be needed for it.

With that configuration, in the event of a Front-end failure, Would the running Virtual Machines still be accesible?

I know that with ssh transfer mode in Filesystem Datastore it is not possible to live-migrate VMs, but, Will it be possible to migrate them if I turn them off? Are there any other ways to migrate VMs between hosts when using Filesystem with ssh?

Sorry to bother you with those newbie questions… I will thank you a lot if you have some time to answer me.

2

Hello,

I’m far from being an experienced OpenNebula user, but I think I can help with some of your questions…

  • 01 single front-end: in a way it depends on what kind of service you’ll be providing (specially regarding how available it needs to be)…
    Loosing the front-end means not being able (for the time taken to recover it) to manage your cloud, both from the cloud admin perspective and from your (possible) self-service portal open to your users. Are you OK with a long recovery time?
    If you can’t risk having your cloud control environment down for a long time, having more than one front-end may be better for you. (The minimum number of servers for front-end HA setup is 3, by the way.)
    One advice: if you planning for (or can only have) a single front-end, pay even more attention on how are you going to backup/restore your configuration and database.

  • I don’t have yet enough experience to comment about capacity planning, sorry.

  • If the front-end goes down, the running VMs will still be accessible. You just won’t be able to use OpenNebula to control them.
    VM(s) per say will only potentially be in trouble if the KVM node hosting them gets in trouble. (There are ways of mitigating this problem by using VM hooks but, again, I’m not there yet…)

  • For non-live migration, you don’t even need to turn them off. They can be migrated even if they are in the running state. OpenNebula will take care of bring them down, migrate, bring them up on the destination node.

I hope I could help,

Alex

3

I really apreciate your help.

There is no problem with recovery times of the front-end. What I was worried about was the posibility of loosing the virtual machines in the event of a front-end failure. But as I understand, it wont happen.

I will follow your advice about the backups.

Another question I have is about networking. Will it be enough with dummy bridge? What will I be missing if I use it?