Hello
I already have 3 Linux servers running VirtualBox hypervisor, is there a way to build an HA cluster (like with qemu-kvm and VMware VSphere) ?
I'm asking for Virtualbox clustering capabilities, not linux clustering capabilities
Thanks for answer
Kind regards
nbanba
VirtualBox cluster
-
- Site Moderator
- Posts: 20965
- Joined: 30. Dec 2009, 20:14
- Primary OS: MS Windows 10
- VBox Version: PUEL
- Guest OSses: Windows, Linux
Re: VirtualBox cluster
I have not used these apps, so I'm not aware what they do. A description of what you want to have happen may guide us better than referring to other apps' functions.nbanba wrote:like with qemu-kvm and VMware VSphere
I believe the closest thing Virtualbox comes to high-availability is Teleporting https://www.virtualbox.org/manual/ch07.html#teleporting I've not done this myself, but it appears to involve storing the VM folder on a central HA network storage, then running the VM on one of two or more identical* physical servers. The teleport settings somehow push the VM off to another of the servers and theoretically keep running if the running server stops.nbanba wrote:is there a way to build an HA cluster
( * "identical": Virtualbox has CPU "leaf" editing commands that limit CPU capabilities so different-CPU-type servers can look identical across their capabilities, so the teleport process can move the VM around. I suspect the OS has to be installed on the lowest-common-denominator CPU limits so it can be teleported.)
Re: VirtualBox cluster
Hello
Thanks answering
I would like to builds an Highly Convergent Infrastructure with 3 identical nodes virtualising the compute part (CPU/RAM),
the network part (beeing able to extend L2 over L3 using VxLAN or GENEVE protocole, to do intra vlan filtering and to automate vlan / vxlan deployment) ,
and the storage part (building a big stack with all storage of the 3 nodes in different tieering and after using things like object storage methode to natively duplicate VM storage chart at file/dir level to 2 of the 3 nodes, transparently for the VM)
The objective is to build a cluster where I can loose a node and all VM of this node will remains availiables from the 2 others with no downtime (with hot replication of session and memory transcluster, like in a firewall HA cluster with pickup session where you can electrically crash a member and all opened sessions remains active with less of 50 millisecond of latency).
It would also be nice to have an inside process which can move VM from one physical node to another to balance the load and use all hardware ressource at the same level
Having a unified interface to manage the cluster where I can add physical node, remove physical node, consume ressources (compute, network and storage) at the cluster level and not at the physical node level, and facilities to clone, export, snapshot, etc.. VMs like it can be done today in Virtualbox on a single node
To be more explicit, having a "Super Virtualbox" which aggregate the 3 physical virtualbox nodes and all VMs in a single interface would be perfect
(like Ovirt does with qemu-kvm and VMWare Vcenter does for VSphere - sorry again for the product name)
I would have a look on "teleporting" to see if it could make the job.
Thanks and kind regards
nbanba
Thanks answering
I would like to builds an Highly Convergent Infrastructure with 3 identical nodes virtualising the compute part (CPU/RAM),
the network part (beeing able to extend L2 over L3 using VxLAN or GENEVE protocole, to do intra vlan filtering and to automate vlan / vxlan deployment) ,
and the storage part (building a big stack with all storage of the 3 nodes in different tieering and after using things like object storage methode to natively duplicate VM storage chart at file/dir level to 2 of the 3 nodes, transparently for the VM)
The objective is to build a cluster where I can loose a node and all VM of this node will remains availiables from the 2 others with no downtime (with hot replication of session and memory transcluster, like in a firewall HA cluster with pickup session where you can electrically crash a member and all opened sessions remains active with less of 50 millisecond of latency).
It would also be nice to have an inside process which can move VM from one physical node to another to balance the load and use all hardware ressource at the same level
Having a unified interface to manage the cluster where I can add physical node, remove physical node, consume ressources (compute, network and storage) at the cluster level and not at the physical node level, and facilities to clone, export, snapshot, etc.. VMs like it can be done today in Virtualbox on a single node
To be more explicit, having a "Super Virtualbox" which aggregate the 3 physical virtualbox nodes and all VMs in a single interface would be perfect
(like Ovirt does with qemu-kvm and VMWare Vcenter does for VSphere - sorry again for the product name)
I would have a look on "teleporting" to see if it could make the job.
Thanks and kind regards
nbanba
-
- Site Moderator
- Posts: 20965
- Joined: 30. Dec 2009, 20:14
- Primary OS: MS Windows 10
- VBox Version: PUEL
- Guest OSses: Windows, Linux
Re: VirtualBox cluster
Hmm, that sounds like quite an ambitious project! Thanks for the further info.
I haven't seen anything in Virtualbox that approaches that level of HA. I'd lean toward making the services the VMs provide being HA rather than the VMs themselves. The same processes that physical servers use to provide HA could then be done in the VMs.
I haven't seen anything in Virtualbox that approaches that level of HA. I'd lean toward making the services the VMs provide being HA rather than the VMs themselves. The same processes that physical servers use to provide HA could then be done in the VMs.
Re: VirtualBox cluster
Hello
Thanks again for answer
You say :
Beeing agnostic of hardware is a part but beeing agnostic of what is executed and providing HA at the hardware level would solved the most part of the problem.
I will let you know when I will have test "teleportinig"
Thanks again
Kind regards
nbanba
Thanks again for answer
You say :
The real major challenge would be to provide the HA at the physical layer level, in order to be agnostic of which technology or which family kind of technology is running inside, and more generally to be agnostic of the front end user execution program or of the inside services of the virtualized machine
I'd lean toward making the services the VMs provide being HA rather than the VMs themselves
Beeing agnostic of hardware is a part but beeing agnostic of what is executed and providing HA at the hardware level would solved the most part of the problem.
I will let you know when I will have test "teleportinig"
Thanks again
Kind regards
nbanba