Hyperconvergence

ullbeking

n00b
Joined
Aug 17, 2017
Messages
32
Hello,

Can somebody please recommend a no-BS introductory tutorial ot the theory and practice of hyperconvergence?

I've searchd but most things I find seem to have been written with marketing in mind or for an enterprise target audience. I"m looking for something a little bit more... real. Does that make sense?

Thanks for any recommendations!!

ubk

Edit: Please nothing /too/ ESX-specific. I'm more of a KVM and Xen kinda guy.
 
Hyperconverged theory = Combined compute and storage in one.

I use Nutanix, in our data center we collapsed 2.5 full racks of SAN and hosts into 12U's, increased performance, decreased power, footprint and management of the entire cluster.
 
Buzzwords. That's it.

And companies selling a single BOM for compute + network + storage + hypervisor.
 
Buzzwords. That's it.

And companies selling a single BOM for compute + network + storage + hypervisor.

Buzzword, yes

Single BOM, No, a flexpod is a single sku but you get a san, storage networking, hosts all separate.

Hyperconvereged consolidates it all into a single appliance.
 
Buzzword, yes

Single BOM, No, a flexpod is a single sku but you get a san, storage networking, hosts all separate.

Hyperconvereged consolidates it all into a single appliance.

But what really defines it as a single appliance? A shared management tool?

It's simply a buzzword from infrastructure vendors with the hopes of selling more gear.

I've run VCE VBlock and Nutanix and it's all just smoke and mirrors...
 
What I usually hear is people want HCI for 3 main reasons:

1. Ease of scale
2. Single pane of glass to manage the entire stack
3. Direct access to storage resources by VM and App Admins, no need for Storage Admins to get in the way

Most of these reasons are directly related to storage. Compared to legacy products like VNX, VMAX, HUS, FAS, DS, etc. where storage management requires training, sometimes specialized staff, and planning for storage resources that ends up being difficult to expand and use, the appeal of "lego blocks" for storage becomes apparent. However, HCI simply moves the burden of storage management to the VM Admin. Do you turn on dedupe? Compression? Encryption? Erasure coding? What RF factor do you choose? What are the affects of those choices on available storage, performance, and availability? What is the right "lego block" to start with for compute, memory, and storage based on the size of your largest, most demanding VMs? How much CPU and memory resources do you sacrifice on each node for the storage layer? How difficult is it to procure and manage as your compute and storage needs grow at different scale and you have a large mix of different nodes, all purchased at different times, with different EOS/EOL?

On top of that, the storage industry has been notorious for selling people a product then smashing them over the head with it 3-5 years later with crazy maintenance bills forcing customers to pay maintenance extortion or replace everything and migrate the data. HCI fixes this by eliminating the latter but embraces the former. A node has a limited lifespan and then you either pay higher maintenance, go 3rd party maintenance (yuck), or replace the node entirely which keeps you in a never ending cycle of buying and replacing your investment both on the compute and storage layer as they're intimately tied together. It's no wonder legacy storage companies are developing, acquiring, and heavily promoting HCI.

The management consideration is a very valid one. Business leaders continue to question why they need one set of experts for storage and another for virtualization. Why can't one set of IT specialists do both and they can divert headcount towards revenue generating or enabling roles like DevOps? But is HCI a requirement for having a simple, elegant, single pane of glass to manage infrastructure?

In my opinion if the storage layer becomes completely simple in scale, management, and performance tuning where no storage admin is required, it integrates seamlessly into the compute layer for "single pane of glass" management, requires no decision making (all data services enabled all the time with 100% performance), and scales with ease both in capacity and performance, and has an open API that supports any orchestration and consumption product of your choosing, doesn't that also solve all 3 of these points even better? You can scale compute and storage independently, the storage tuning and management requires next to zero FTE and decision points, and you can automate and enable self provisioning with the toolkit of your choice.

HCI positions itself as a solution for the three points above but I'd argue it's not the best nor is it the only solution. But HCI vendors (both new and legacy storage companies peddling it) would have you believe otherwise.
 
  • Like
Reactions: kdh
like this
^^^ 100% Agree. Very much on point.

We have a flexpod here that works ok, and I'm looking at EMC's offering at the moment. Its not so much buzz words as the landscape hasn't fully matured just yet. 5 or 6 years ago, it was very duct tape and bolted together.. Gen 1 Vblock comes to mind. But a lot of the big players have finally come to the table to make a solid solution. However, one thing to consider, if you already have storage, network, servers and a hypervisor of sorts, there isn't much reason to look a HCI infrastructure unless you are completely refreshing your current platform or all your infrastructure guys left the organization and you are starting over.

@OP, KVM and Zen support will be minimal at the moment. VMware being the 800 pound monster seems to get all the attention, but I see that changing soon. But the jist of HCI is a known stack of hardware, software with a single management interface to make it all work. You can't update and shouldn't manage either piece of the hardware individually, but have to manage it from the management software provided, and deploy an update to the whole stack that has been pre-vendor approved. There are positives and negatives to this approach, meaning, you know you have a stack that is known good or, you have a platform breaking bug that can't be fixed until the whole HCI stack gets an update. Also HCI stacks between vendors don't necessarily mean the same thing. You also need to know your work load. Are you heavily random? Reads/Writes? Sequential? Most HCI solutions are designed for specific work loads. Data warehouse, OLTP, VMware, or some other work load. So you have to know the use case before looking at an HCI solution. If you do not have a defined use case, you are kind of wasting yours and the vendors time.

HCI is very cost prohibitive to the average user unless you design your own cobbled together solution, hence why its aimed at the enterprise these days. Can I recommend a platform? No. There are so many players in the market that just sprang up over night that its very fly by night. However, most of them will fold or get bought out by the big players anyway.. Cisco, EMC, HP, and Dell...hell.. Even Oracle's Exadata platform. Hopefully that helps. best of luck.
 
Really think of HCI as a vendor all-in-one-package deal with that single pane of glass for management. A lot of vendors are going the openstack and REST api route for interfacing with their gear nowadays.

HCI has a lot of options now with buzzword mania all over. Take your time to learn. If you're buying anything, don't let pushy sales guys try to take advantage of you. Get some trial gear in house to give it a shot if you like how something looks. Sometimes the single pane of management can be more of a pain when you realize components don't play nicely with how some solutions are legoed together (looking at you early VCE shit that really wasnt a hyperconverged solution).

As you keep digging please come ask more questions, I've found a number of vendor/ex-vendor people in these forums that can provide a bit more of a non-BS answer to specific solutions. The extra few minutes of your time may save you hours of bullshit later on down the road

Definitely go digging more into openstack and vendor neutral APIs that are more popular. You'll find that the demand from cloud providers has been so high that vendors are rolling their own neutral drivers/ control paths to cater sales to everyone and get a piece of the pie
 
Back
Top