Infrastructure Needed for Hypervisor Cluster
-
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
When I looked at it I came to the conclusion that I would need a minimum of three nodes and they should be the same CPU generation and have the same NIC configuration.
That's a VMware requirement, not a clustering requirement. And even on VMware, it's not an actual requirement, they just need a third witness that isn't part of the cluster.
But what most people do is put that Witness on their 2-node cluster. . . .
Which could cause issues for all kinds of reasons. .
-
@DustinB3403 said in Infrastructure Needed for Hypervisor Cluster:
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
When I looked at it I came to the conclusion that I would need a minimum of three nodes and they should be the same CPU generation and have the same NIC configuration.
That's a VMware requirement, not a clustering requirement. And even on VMware, it's not an actual requirement, they just need a third witness that isn't part of the cluster.
But what most people do is put that Witness on their 2-node cluster. . . .
Which could cause issues for all kinds of reasons. .
Yeah, but no need for the witness node outside of VMware.
-
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@DustinB3403 said in Infrastructure Needed for Hypervisor Cluster:
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
When I looked at it I came to the conclusion that I would need a minimum of three nodes and they should be the same CPU generation and have the same NIC configuration.
That's a VMware requirement, not a clustering requirement. And even on VMware, it's not an actual requirement, they just need a third witness that isn't part of the cluster.
But what most people do is put that Witness on their 2-node cluster. . . .
Which could cause issues for all kinds of reasons. .
Yeah, but no need for the witness node outside of VMware.
I think you mean, no need for a 3rd ESXi VMWare box just to run the node.
It is recommended to be outside of the cluster, but they specifically say it does work if you wanted it on the cluster. (with a crap ton of asterisks following the statement)
-
@DustinB3403 said in Infrastructure Needed for Hypervisor Cluster:
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@DustinB3403 said in Infrastructure Needed for Hypervisor Cluster:
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
When I looked at it I came to the conclusion that I would need a minimum of three nodes and they should be the same CPU generation and have the same NIC configuration.
That's a VMware requirement, not a clustering requirement. And even on VMware, it's not an actual requirement, they just need a third witness that isn't part of the cluster.
But what most people do is put that Witness on their 2-node cluster. . . .
Which could cause issues for all kinds of reasons. .
Yeah, but no need for the witness node outside of VMware.
I think you mean, no need for a 3rd ESXi VMWare box just to run the node.
It is recommended to be outside of the cluster, but they specifically say it does work if you wanted it on the cluster. (with a crap ton of asterisks following the statement)
Correct, can be on a desktop or whatever.
-
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
I don't think two nodes is enough if you want to play with clusters. Better to have more nodes with less ram/cpu and storage. Like 4 or 6 or something.
Not only is it enough, it's often recommended. There is no such requirement of needing three servers.
Xenserver recommends three (or more) for HA.
-
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
I don't think two nodes is enough if you want to play with clusters. Better to have more nodes with less ram/cpu and storage. Like 4 or 6 or something.
Not only is it enough, it's often recommended. There is no such requirement of needing three servers.
Xenserver recommends three (or more) for HA.
Everyone recommends 3 or more hosts.
-
@DustinB3403 said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
I don't think two nodes is enough if you want to play with clusters. Better to have more nodes with less ram/cpu and storage. Like 4 or 6 or something.
Not only is it enough, it's often recommended. There is no such requirement of needing three servers.
Xenserver recommends three (or more) for HA.
Everyone recommends 3 or more hosts.
Not everyone. Starwinds does not, for example.
-
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
I don't think two nodes is enough if you want to play with clusters. Better to have more nodes with less ram/cpu and storage. Like 4 or 6 or something.
Not only is it enough, it's often recommended. There is no such requirement of needing three servers.
Xenserver recommends three (or more) for HA.
XS is one specific vendor that sells licenses. So they have a financial incentive to push the 3+ host model. And their standard shared storage mechanism only works well with 3+. So they chose a system based around that, so that they recommend it is expected.
But products like DRBD and Starwind are engineered all around the 2 node model. They scale up, but their intent is focused on two nodes.
-
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
XS is one specific vendor that sells licenses. So they have a financial incentive to push the 3+ host model. And their standard shared storage mechanism only works well with 3+. So they chose a system based around that, so that they recommend it is expected.
But products like DRBD and Starwind are engineered all around the 2 node model. They scale up, but their intent is focused on two nodes.
HPE Simplivity works normally with 2 nodes so does Datacore's VSAN. Although both require/should have a witness. I was looking into Simplicity to run a vSphere cluster some time ago but after HPE had bought them their support hit rock bottom at some stage. DataCore is a bit of a dark sheep, I wonder if anyone on ML ever used them at all. I'd look at Starwind and VMware VSANs first, both have large vendor and IT community support. Unless the OP wants to do some KVM with DRBD, Gluster, Ceph etc.
-
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
Clustering is done when the cost of clustering is low versus the risk of not clustering.
I would change that slightly to Clustering is done when the cost of clustering is low versus the cost of not clustering. A risk is always a cost, but some costs are not risks. For example taking down a hypervisor for maintenance vs moving guests to another node and taking down a free node for maintenance.
-
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@DustinB3403 said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
@Pete-S said in Infrastructure Needed for Hypervisor Cluster:
I don't think two nodes is enough if you want to play with clusters. Better to have more nodes with less ram/cpu and storage. Like 4 or 6 or something.
Not only is it enough, it's often recommended. There is no such requirement of needing three servers.
Xenserver recommends three (or more) for HA.
Everyone recommends 3 or more hosts.
Not everyone. Starwinds does not, for example.
fair enough
-
@Vlinderbeest said in Infrastructure Needed for Hypervisor Cluster:
@scottalanmiller said in Infrastructure Needed for Hypervisor Cluster:
Clustering is done when the cost of clustering is low versus the risk of not clustering.
I would change that slightly to Clustering is done when the cost of clustering is low versus the cost of not clustering. A risk is always a cost, but some costs are not risks. For example taking down a hypervisor for maintenance vs moving guests to another node and taking down a free node for maintenance.
Sure, using risk cost vs investment cost as cost v cost is a perfect valid way to look at it. I use that all the time in the opposite way.
You can say that risk is a cost. Or conversely, you can look at the clustering cost up front as essentially a "financial event" similar to an outage.
