hyper-converged explained without the hype

[morpheus]

Well, the video quite explains the comparation between tradicional architecture and Hyper converged infrastructure (HCI). There are some assumptions that are in part accurate.
HCI systems does not need to grow in compute anytime you want to grow in storage. One of the main advantages of HCI system is that you can Scale up (add disks to existing nodes) of scale out (add nodes, thus compute+storage). Obviously to scale up you need to have free disk slots in your servers.


On the other hand, storage array (at least midrage ones) are compound by 2 controllers, and you cannot add more. So the only way to grow is adding disks in new enclosures, but you are limited to 2 controllers. High range arrays (Dell PowerMax ie.) can grow bot in disks and controllers.


About networking, well, the bottleneck would be your switch not the HCI nodes. HCI nodes already can use 25GbE NICs, so you can aggregate and create a 100GbE link to your Top Of Rack switch. Don't think the network could suppose an issue here, although it is true that has to be carefully configured to get all the performance.


To finish, HCI is not intented for all kind of workloads. It works great for VDI environments, migration from already virtualized enviroments, even big data, but it is true that it not intented for extreme compute (HPC) or huge data set processing.


Regarding to the cost, you are having into account only CAPEX, but you are not talking about operations (OPEX). HCI systems are MUCH easier to manage, to set and the lifecycle management it quite simple. So the provisioning and service delivery is faster and more granular, saving time and reducing error posibility, thus saving OPEX.

[shanegoh]

Best explanation I've come across. Thanks. The animations really help as well.

[drewzhao]

Great. The disadvantages explained are helpful.

[amrkoptan]

great video, simple and straight forward, thanks a lot

[Juicy_J]

Correction: in DellEMC’s VxRail HCI Solution you can grow storage without growing compute. There are typically many drive slots available across a 4 node cluster. You can also grow compute without growing storage. When both need to grow you add a node.

[XLinkBizTech]

Excellent, short, precise and to the point. Thanks.

[RadekPogoda]

Love the simple, but not simplistic approach and delivery. Great job!

[nirajshah]

thank you so much to keep it short and straight

[ychilds]

Thank you for mentioning some limitations!

[Athree]

Explain in neutrality. Well done. Thanks.

[Nikoolayy]

So we are returning to mainframes :), nothing new here. A device with local storage, CPU, memory and a network switch.

[teaearlgrayht]

You need network switches even with hyper

[angelg]

My colleagues want to marry with DellEMC and VxRail. Is it a good idea, considering that it is NOT commodity hardware and we will become locked-in to a single vendor ? Also HCI doesn't look as a greener technology given that the CPUs which used to be offloaded from PIO IOPS (programmed input-output), now will be tasked to convert every single I/O into TCP/IP packets to make possible the software-defined storage of the HCI (unless vendors add a proprietary magical black box to act as a filesystem accelerator).

[notchcrafter]

The "But's" are not all true.

[GouriDas]

Thoroughly enjoyed this. Thanks. Much appreciated.

[gride]

How do Nutanix stack up in this group now?

[fbifido]

@1:28 - Networking is not included in that package.

[MyWaloo]

Thank you very match , for the presentation and the advice