Hyper-Converged Infrastructure

[Jeff7181]

HCI is an aging buzzword (it feels like it's been a couple years since it was "cool" to be heard conversing about "hyper-converged"). But I've noticed products are being refined and the big players are starting to adopt and release their own HCI.

So my question is: an obvious use case is a large environment where storage and compute scale mostly linearly. But what about smaller environments where they don't scale so linearly? What about a branch office that doesn't have very high demand for compute or network resources, but needs maybe 100TB of storage? Is HCI still a potential fit even if you only need 4-8 CPU cores, less than 32 GB of RAM and don't need a bunch of 10 GbE ports?

 

[XavierMace]

The whole point of HCI is bringing all the parts of your infrastructure into one centralized software management system. A branch office isn't really gaining anything as they don't have much hardware to begin with. If you've only got one server with onboard storage, there's not anything to converge (or hyperconverge). Things like VMWare vSAN requires more hardware than they have.

 

[Jeff7181]

The gain in that situation wouldn't necessarily be centralized management, but unified management.

My question is more in regard to how HCI seems to scale mostly linearly. Are there vendors with HCI nodes that are compute heavy, or network heavy, or storage heavy? I haven't seen any, but that doesn't mean they don't exist.

That used to be one of the big selling points of Isilon in the scale-out NAS market. They had performance nodes, storage nodes and combo nodes and they could be added to the cluster in any combination at any time.

 

[[DHT]Osiris]

A lot of the software solutions (including MS storage pools now) can accept just a blob of storage in whatever form you can deliver it, either iSCSI through a switch to a FC, FCoE, iSCSI, 10GBASE-T, or coax array, either JBOD or some sort of super-storage-awesomeness NAS system. In addition, once you get enough of this stuff together they can start doing parity across the systems, so you can get creative with what storage you're using, where it's located, etc.

For your specific query, you can craft a vm environment with 'compute nodes' and 'storage nodes' pretty easily. Last place I was at used a Cisco blade system which was basically entirely compute-focused for the running of the VMs, attached to netapp storage arrays for the actual storage. So it'd be a self-contained rack with 4-8 cisco blades in a 8-blade chassis (standardized across all sites), connected to a fat array of disks (also standardized). We had everything separated by NAS (so LUNs/partitions/whatevers didn't cross NAS') but you could easily have just basically turned this into a giant superblob of storage and just let VMWare vSAN or hyper-v storage pools handle the mess.

 

[dave_the_nerd]

HCI is an aging buzzword (it feels like it's been a couple years since it was "cool" to be heard conversing about "hyper-converged"). But I've noticed products are being refined and the big players are starting to adopt and release their own HCI.

Yes.

So my question is: an obvious use case is a large environment where storage and compute scale mostly linearly. But what about smaller environments where they don't scale so linearly? What about a branch office that doesn't have very high demand for compute or network resources, but needs maybe 100TB of storage? Is HCI still a potential fit even if you only need 4-8 CPU cores, less than 32 GB of RAM and don't need a bunch of 10 GbE ports?

Depends on which vendor you're working with and how their "HCI" solution is designed. Some scale like you're describing, others are built around tech that can be scaled one way without forcing the other. At least I'm pretty sure those products have gotten to market. No comment.

And it's the job of a good IT department to know which buzzwords do or do not fit their use case, and when. In a lot of circumstances, the "old fashioned way" with distinct compute and storage resources may still make sense.