What makes a system HCI?
-
Terminology often changes as time passes on. I don't think there is a strict definition for these terms.
Also, I don't think "Hyper" in hyperconverged has anything to do with hypervisor. It's just another word for "very much" or "super" or "ultra".
So HCI and CI are very much alike except that hyper is even more. So for example if you have a bunch of hypervisors that connect to a SAN you have CI. But if you instead of a SAN do a vSAN on the hypervisors you have HCI.
So on HCI I think most people agree that you need to have:
- Compute virtualization
- Networking virtualization
- Storage virtualization
-
That does make total sense. One discussion staff keep having internally is that HCAs from vendors have 1 x NIC only. Therefore, if a server has 2 x NIC, or more, it cannot be HCI... which I think is total bull.
-
To put this simply, every server that has compute and storage in the box is hyperconverged.
Buy* any server off of a shelf that is "self-contained" and you have a hyperconverged server. The value add in an HCI solution is the programming that allows you to take 2,3,4 or more of those servers and just plug them in and use it all as one large server pool (HCI).
So any host in the environment could go down, and while you'd have reduced capacity, the environment would simply move the workload to other available resources.
-
@Jimmy9008 said in What makes a system HCI?:
That does make total sense. One discussion staff keep having internally is that HCAs from vendors have 1 x NIC only. Therefore, if a server has 2 x NIC, or more, it cannot be HCI... which I think is total bull.
What sort of illogic led to the number of anything, let alone # of NICs, in a server being HCI or not?
-
@DustinB3403 said in What makes a system HCI?:
To put this simply, every server that has compute and storage in the box is hyperconverged.
Buy* any server off of a shelf that is "self-contained" and you have a hyperconverged server. The value add in an HCI solution is the programming that allows you to take 2,3,4 or more of those servers and just plug them in and use it all as one large server.
So any host in the environment could go down, and while you'd have reduced capacity, the environment would simply move the workload to other available resources.
That is what I have always in my mind for HCI. Am I right in saying the value add can be done by say a Windows Failover Cluster over all nodes, which make use of vSAN storage? Like the HCA appliances vendors sell, the failover cluster provides the function to move VMs should a node fail, right? The vendor tech isnt some magic box which invalidates other solutions excluding such technology from being HCI?
-
@travisdh1 said in What makes a system HCI?:
@Jimmy9008 said in What makes a system HCI?:
That does make total sense. One discussion staff keep having internally is that HCAs from vendors have 1 x NIC only. Therefore, if a server has 2 x NIC, or more, it cannot be HCI... which I think is total bull.
What sort of illogic led to the number of anything, let alone # of NICs, in a server being HCI or not?
Simply, when they are looking at HCA from vendors, say Nutanix, Dell, VMWare, Scale, the manual appears to have 1 x NIC in each node, which has virtualized storage network, VM network, heartbeats and other such networks on top of the one NIC using different vLANs. I disagree with them entirely, but its at a point where any architecture using more than one NIC in their mind cannot be HCI.
-
@Jimmy9008 said in What makes a system HCI?:
@DustinB3403 said in What makes a system HCI?:
To put this simply, every server that has compute and storage in the box is hyperconverged.
Buy* any server off of a shelf that is "self-contained" and you have a hyperconverged server. The value add in an HCI solution is the programming that allows you to take 2,3,4 or more of those servers and just plug them in and use it all as one large server.
So any host in the environment could go down, and while you'd have reduced capacity, the environment would simply move the workload to other available resources.
That is what I have always in my mind for HCI. Am I right in saying the value add can be done by say a Windows Failover Cluster over all nodes, which make use of vSAN storage? Like the HCA appliances vendors sell, the failover cluster provides the function to move VMs should a node fail, right? The vendor tech isnt some magic box which invalidates other solutions excluding such technology from being HCI?
Windows Failover Clustering isn't seemless (you'd know if you lost a host). vSAN is storage only (generally). @scale makes HCI environments where you can take a host right out of the environment and things will just chug along.
As for your second question I'm not sure what you're asking regarding vendor tech being magic. Magic is simply something we don't understand yet.
-
@Jimmy9008 said in What makes a system HCI?:
I disagree with them entirely, but its at a point where any architecture using more than one NIC in their mind cannot be HCI.
Who is seeing any HCI vendor sell equipment with only 1 physical NIC on each system?
-
@DustinB3403 said in What makes a system HCI?:
@Jimmy9008 said in What makes a system HCI?:
@DustinB3403 said in What makes a system HCI?:
To put this simply, every server that has compute and storage in the box is hyperconverged.
Buy* any server off of a shelf that is "self-contained" and you have a hyperconverged server. The value add in an HCI solution is the programming that allows you to take 2,3,4 or more of those servers and just plug them in and use it all as one large server.
So any host in the environment could go down, and while you'd have reduced capacity, the environment would simply move the workload to other available resources.
That is what I have always in my mind for HCI. Am I right in saying the value add can be done by say a Windows Failover Cluster over all nodes, which make use of vSAN storage? Like the HCA appliances vendors sell, the failover cluster provides the function to move VMs should a node fail, right? The vendor tech isnt some magic box which invalidates other solutions excluding such technology from being HCI?
Windows Failover Clustering isn't seemless (you'd know if you lost a host). vSAN is storage only (generally). @scale makes HCI environments where you can take a host right out of the environment and things will just chug along.
As for your second question I'm not sure what you're asking regarding vendor tech being magic. Magic is simply something we don't understand yet.
I understand. So, since WFC is not seamless if a host is removed, that means its not HCI? Not arguing, just trying to understand where the line is.
So, if removing a node != seamless, then solution != HCI?
-
@DustinB3403 said in What makes a system HCI?:
@Jimmy9008 said in What makes a system HCI?:
I disagree with them entirely, but its at a point where any architecture using more than one NIC in their mind cannot be HCI.
Who is seeing any HCI vendor sell equipment with only 1 physical NIC on each system?
They have had solutions like VXRail on trial to see how they work from various vendors, each node in the stack only has 1 NIC card. Maybe they are just entry level systems or something.
-
@DustinB3403 said in What makes a system HCI?:
@Jimmy9008 said in What makes a system HCI?:
@DustinB3403 said in What makes a system HCI?:
To put this simply, every server that has compute and storage in the box is hyperconverged.
Buy* any server off of a shelf that is "self-contained" and you have a hyperconverged server. The value add in an HCI solution is the programming that allows you to take 2,3,4 or more of those servers and just plug them in and use it all as one large server.
So any host in the environment could go down, and while you'd have reduced capacity, the environment would simply move the workload to other available resources.
That is what I have always in my mind for HCI. Am I right in saying the value add can be done by say a Windows Failover Cluster over all nodes, which make use of vSAN storage? Like the HCA appliances vendors sell, the failover cluster provides the function to move VMs should a node fail, right? The vendor tech isnt some magic box which invalidates other solutions excluding such technology from being HCI?
Windows Failover Clustering isn't seemless (you'd know if you lost a host). vSAN is storage only (generally). @scale makes HCI environments where you can take a host right out of the environment and things will just chug along.
As for your second question I'm not sure what you're asking regarding vendor tech being magic. Magic is simply something we don't understand yet.
Oh, missed the second part. By magic I mean the vendor talk where I keep hearing you just plug in another HCA unit to expand resources. You don't need to understand whats happening under the hood, you just need to drop some cash and plug the next node in to the HCA to grow. That is what I meant by magic.
I get that a vendor has some cool tech they stick on top of their HCI hardware to sell me a HCA, but you can still have HCI without that super cool layer on top that they have, right? Or are we saying HCI can only ever be HCI if it has all the bells on top that vendors sell through their proprietary software/stack?
-
@Jimmy9008 said in What makes a system HCI?:
@travisdh1 said in What makes a system HCI?:
@Jimmy9008 said in What makes a system HCI?:
That does make total sense. One discussion staff keep having internally is that HCAs from vendors have 1 x NIC only. Therefore, if a server has 2 x NIC, or more, it cannot be HCI... which I think is total bull.
What sort of illogic led to the number of anything, let alone # of NICs, in a server being HCI or not?
Simply, when they are looking at HCA from vendors, say Nutanix, Dell, VMWare, Scale, the manual appears to have 1 x NIC in each node, which has virtualized storage network, VM network, heartbeats and other such networks on top of the one NIC using different vLANs. I disagree with them entirely, but its at a point where any architecture using more than one NIC in their mind cannot be HCI.
Wow, just wow.
Scale systems come with 4 NICs by default. A base config was 2x10Gb for the storage layer and 2x1Gb for eveything else.
If VXRail is only using a single NIC for everything, no wonder their base configuration is so bad!
-
Just had a look on the VXRail spec pages and that also allows for expanding network cards for capacity/failures. So that helps my argument at least.
-
@Jimmy9008 said in What makes a system HCI?:
I get that a vendor has some cool tech they stick on top of their HCI hardware to sell me a HCA, but you can still have HCI without that super cool layer on top that they have, right? Or are we saying HCI can only ever be HCI if it has all the bells on top that vendors sell through their proprietary software/stack?
You could do HCI yourself, sure but building the tools to get it aren't something any individual would reasonably do.
-
@travisdh1 said in What makes a system HCI?:
@Jimmy9008 said in What makes a system HCI?:
@travisdh1 said in What makes a system HCI?:
@Jimmy9008 said in What makes a system HCI?:
That does make total sense. One discussion staff keep having internally is that HCAs from vendors have 1 x NIC only. Therefore, if a server has 2 x NIC, or more, it cannot be HCI... which I think is total bull.
What sort of illogic led to the number of anything, let alone # of NICs, in a server being HCI or not?
Simply, when they are looking at HCA from vendors, say Nutanix, Dell, VMWare, Scale, the manual appears to have 1 x NIC in each node, which has virtualized storage network, VM network, heartbeats and other such networks on top of the one NIC using different vLANs. I disagree with them entirely, but its at a point where any architecture using more than one NIC in their mind cannot be HCI.
Wow, just wow.
Scale systems come with 4 NICs by default. A base config was 2x10Gb for the storage layer and 2x1Gb for eveything else.
If VXRail is only using a single NIC for everything, no wonder their base configuration is so bad!
Hey they need a way to upsell
-
@Jimmy9008 said in What makes a system HCI?:
Oh, missed the second part. By magic I mean the vendor talk where I keep hearing you just plug in another HCA unit to expand resources.
The last time I did this with @scale , it worked just like that. Plug the new server in, tell the other systems where to find the new server, and off to the races you go.
-
@dafyre said in What makes a system HCI?:
@Jimmy9008 said in What makes a system HCI?:
Oh, missed the second part. By magic I mean the vendor talk where I keep hearing you just plug in another HCA unit to expand resources.
The last time I did this with @scale , it worked just like that. Plug the new server in, tell the other systems where to find the new server, and off to the races you go.
Yeah, that's a part of the tooling they've built to make "Scale". Could someone else maybe build the same thing, sure but at what cost?
-
@dafyre said in What makes a system HCI?:
@Jimmy9008 said in What makes a system HCI?:
Oh, missed the second part. By magic I mean the vendor talk where I keep hearing you just plug in another HCA unit to expand resources.
The last time I did this with @scale , it worked just like that. Plug the new server in, tell the other systems where to find the new server, and off to the races you go.
I have no doubt this is true. None at all. I am in no way saying the system cannot do this functionality.
What I am trying to get an insight in to is..... if the system does not do the above, does that mean the system is not HCI?
-
@Jimmy9008 said in What makes a system HCI?:
if the system does not do the above, does that mean the system is not HCI?
No, one does not mean that the other HCI solutions aren't HCI. It just means that the tooling isn't there / included.
Different HCI solutions can have different features.
-
@DustinB3403 said in What makes a system HCI?:
@Jimmy9008 said in What makes a system HCI?:
I get that a vendor has some cool tech they stick on top of their HCI hardware to sell me a HCA, but you can still have HCI without that super cool layer on top that they have, right? Or are we saying HCI can only ever be HCI if it has all the bells on top that vendors sell through their proprietary software/stack?
You could do HCI yourself, sure but building the tools to get it aren't something any individual would reasonably do.
Would users of Starwind vSAN, running a three node setup using their vSAN with WFC on top be HCI? Three nodes, all shared storage, to a Windows Failover Cluster running over all three nodes... sure, its not as polished as the scale solutions (never said it is).... but does that mean it is not HCI?
How can this be when right at the start somebody said HCI is:
Compute virtualization
Networking virtualization
Storage virtualization
