Is HyperConvergence Even a Thing
-
So about a week ago I was talking to two technical sales guys about hyperconvergence. Both have work experience in the bench space, but not in IT, but on the relatively high end of the bench space. They were asking how hyperconvergence was special, so we talk about it for a while.
At the end of the discussion, the real takeaway was that hyperconvergence, to them, was just the "default, obvious way to do things." There wasn't any specific special thing about it. And it's true. I think that the industry has become so focused on certain expensive, complex approaches to things, that sometimes going back to basics seems like a special task in and of itself. This was certainly what caused the SAM-SD to be so important, not that it isn't was so important, but that it forced people to shake off false assumptions and go back to the basics.
In reality, every single standalone server is essentially hyperconverged. Through Hyper-V, VMware ESXi, XenServer or KVM on a single node using local storage and that's hyperconverged! The logical way to grow that is to simply scale out what is there... hyperconverged again.
Of course, in the "real" world, hyperconvergence vendors focus on adding special sauce in one form or another rather than just repeating the basics. Whether it is special storage that isn't available in other forms, management tool stacks or whatever. But at its core, HC is such an obvious and intuitive technical step that even to people outside of IT looking in, it's just what they assumed would be done from the beginning.
Stepping back and realizing that we've been using HC since the very first virtual server was built in 1964 and that everything that isn't HC is the newer, less obvious approach is useful. Whether we have one server, or many, HC is the obvious starting point. Whether we want HA or not, HC is the obvious starting point. Sure, there are cases for other things, but they are the exceptions, not the rules.
-
I was of that a hyperconverged setup blurs the line between one node or several (like @scale does)...
But looking at it from this standpoint makes me realize that sometimes, going back to the basics isn't really a bad thing.
-
@dafyre said in Is HyperConvergence Even a Thing:
I was of that a hyperconverged setup blurs the line between one node or several (like @scale does)...
But looking at it from this standpoint makes me realize that sometimes, going back to the basics isn't really a bad thing.
Yeah, that is what Hyperconvergence (in the modern) sense means, so there is no single node where something may be running at. That at a moments notice it can be anywhere else.
The Special Sauce comes in at that point, as really all the hardware is, is enterprise grade equipment engineered with the task at hand in mind.
-
@dafyre said in Is HyperConvergence Even a Thing:
I was of that a hyperconverged setup blurs the line between one node or several (like @scale does)...
But... what changes between one node and two or three? If the tech is the same, what makes one hyperconverged and one not? What isn't hyperconverged about a single node?
-
@DustinB3403 said in Is HyperConvergence Even a Thing:
Yeah, that is what Hyperconvergence (in the modern) sense means, so there is no single node where something may be running at.
I've never seen that in any definition. That's the definition of HA, not HC.
-
@DustinB3403 said in Is HyperConvergence Even a Thing:
Yeah, that is what Hyperconvergence (in the modern) sense means, so there is no single node where something may be running at. That at a moments notice it can be anywhere else.
This is demonstrably not true if you look at many HC solutions. They might allow for this use case, but it is not guaranteed. At best this is a matter of capacity planning, not the technology.
-
@scottalanmiller said in Is HyperConvergence Even a Thing:
@DustinB3403 said in Is HyperConvergence Even a Thing:
Yeah, that is what Hyperconvergence (in the modern) sense means, so there is no single node where something may be running at.
I've never seen that in any definition. That's the definition of HA, not HC.
A hyperconverged solution usually has more than a single node, and while it doesn't require it. If someone is looking to buy a hyperconverged system, they are looking for a node-n setup.
Not a single server, at my last job it was totally a hyperconverged setup, but without the HA aspect. SA was what it was setup for.
2 Servers, 1 running production, the other sitting there as a hot standby receiving CR jobs every 15 minutes.
All components contained individually in each server, but each server able to operate the business needs at a moments notice.
-
@DustinB3403 said in Is HyperConvergence Even a Thing:
@scottalanmiller said in Is HyperConvergence Even a Thing:
@DustinB3403 said in Is HyperConvergence Even a Thing:
Yeah, that is what Hyperconvergence (in the modern) sense means, so there is no single node where something may be running at.
I've never seen that in any definition. That's the definition of HA, not HC.
A hyperconverged solution usually has more than a single node, and while it doesn't require it. If someone is looking to buy a hyperconverged system, they are looking for a node-n setup.
Not a single server, at my last job it was totally a hyperconverged setup, but without the HA aspect. SA was what it was setup for.
2 Servers, 1 running production, the other sitting there as a hot standby receiving CR jobs every 15 minutes.
All components contained individually in each server, but each server able to operate the business needs at a moments notice.
What you're describing here is a failover type setup in my mind.
-
@scottalanmiller said in Is HyperConvergence Even a Thing:
@dafyre said in Is HyperConvergence Even a Thing:
I was of that a hyperconverged setup blurs the line between one node or several (like @scale does)...
But... what changes between one node and two or three? If the tech is the same, what makes one hyperconverged and one not? What isn't hyperconverged about a single node?
With a single node, you are limited to a single node.
I see hyperconvergence as taking multiple nodes and making them closer to behaving more like one node (again, like Scale does). Of course there are limits on each node, but the special sauce (yes, I'm stealing that term) ties the nodes together in such a way that some of these limits can be worked around.
-
Hyperconverged to me is bringing disparate interfaces and technologies into one box. Storage, Hypervisor, Management, and Network on a single box managed from a single interface.
-
@dafyre said in Is HyperConvergence Even a Thing:
I see hyperconvergence as taking multiple nodes and making them closer to behaving more like one node (again, like Scale does).
But it's not. Using definitions like that, we are back to SAN being HC, which isn't the case.
-
@coliver said in Is HyperConvergence Even a Thing:
Hyperconverged to me is bringing disparate interfaces and technologies into one box. Storage, Hypervisor, Management, and Network on a single box managed from a single interface.
That's what it has always been and the only thing that makes sense as a term.
-
@dafyre said in Is HyperConvergence Even a Thing:
@DustinB3403 said in Is HyperConvergence Even a Thing:
@scottalanmiller said in Is HyperConvergence Even a Thing:
@DustinB3403 said in Is HyperConvergence Even a Thing:
Yeah, that is what Hyperconvergence (in the modern) sense means, so there is no single node where something may be running at.
I've never seen that in any definition. That's the definition of HA, not HC.
A hyperconverged solution usually has more than a single node, and while it doesn't require it. If someone is looking to buy a hyperconverged system, they are looking for a node-n setup.
Not a single server, at my last job it was totally a hyperconverged setup, but without the HA aspect. SA was what it was setup for.
2 Servers, 1 running production, the other sitting there as a hot standby receiving CR jobs every 15 minutes.
All components contained individually in each server, but each server able to operate the business needs at a moments notice.
What you're describing here is a failover type setup in my mind.
Exactly. Which might be his point... end users use HC to mean HA. But HC itself isn't HA nor does it imply it.
-
@scottalanmiller said in Is HyperConvergence Even a Thing:
@dafyre said in Is HyperConvergence Even a Thing:
@DustinB3403 said in Is HyperConvergence Even a Thing:
@scottalanmiller said in Is HyperConvergence Even a Thing:
@DustinB3403 said in Is HyperConvergence Even a Thing:
Yeah, that is what Hyperconvergence (in the modern) sense means, so there is no single node where something may be running at.
I've never seen that in any definition. That's the definition of HA, not HC.
A hyperconverged solution usually has more than a single node, and while it doesn't require it. If someone is looking to buy a hyperconverged system, they are looking for a node-n setup.
Not a single server, at my last job it was totally a hyperconverged setup, but without the HA aspect. SA was what it was setup for.
2 Servers, 1 running production, the other sitting there as a hot standby receiving CR jobs every 15 minutes.
All components contained individually in each server, but each server able to operate the business needs at a moments notice.
What you're describing here is a failover type setup in my mind.
Exactly. Which might be his point... end users use HC to mean HA. But HC itself isn't HA nor does it imply it.
Neither did my description of it
-
@scottalanmiller said in Is HyperConvergence Even a Thing:
@dafyre said in Is HyperConvergence Even a Thing:
I see hyperconvergence as taking multiple nodes and making them closer to behaving more like one node (again, like Scale does).
But it's not. Using definitions like that, we are back to SAN being HC, which isn't the case.
But SAN is not like Scale. It only does storage. The Hyperconverged idea brings everything (virtualization, storage, networking, etc) under a single console...whether one node or 10. (Ha! I managed to not say single pane of glass, lol)...
-
-
@coliver said in Is HyperConvergence Even a Thing:
Hyperconverged to me is bringing disparate interfaces and technologies into one box. Storage, Hypervisor, Management, and Network on a single box managed from a single interface.
That's just boring virtualization. (Virtualizing networking, storage, and compute of a local server).
Hyperconverged infrastructure is an evolved term from converged infrastructure.
It was introduced to argue that CI systems were made of disparate vendor solutions and bringing it all together into a system where the hosts compute and disk bays took on the role of the storage controller in providing HA storage. The SAN component was rolled inside the server It was "moar" converged than a traditional CI stack.
Now some people are abusing this term. Netapp is describing a bundled deployment of SolidFire and SuperMicro servers is HCI.
Coheisity is arguing that they are hyperconvered secondary storage (because they converge secondary storage roles).
for a while Gartner said HCI had to be an appliance.
They are/where all wrong.
-
@John-Nicholson then there has to be a set of descriptive terms for the items you're describing.
Multiple SAN's tied together to act as a HA SAN doesn't make it Hyperconverged. Because the system as designed is only working at the storage layer/network layers. There would be no hypervisor or processing power there.
-
@John-Nicholson This makes me wonder what D&D alignment @scottalanmiller would be. Chaotic good?
-
HC eh... What does HC multi box look like if not HA'ed?
For example, I'm not sure I consider MS's hyper-v manger a HC interface even though I'm sure you can put many hyper-v hosts into it so you have a 'single pane of glass.'
If you have 3 Hyper-v or ESXi or whatever all with their own storage, yet managed by a single console are you really HC?
