When Cloud Ate Hyperconverged

Last week, I wrote a blog post "Why I've Lost Interest In Hyperconverged".

My argument, in a nutshell, was that the central value proposition for hyperconverged was taking cost out of infrastructure by consolidating less-important applications.

That creates two strategic problems for its vendors.

First, as it's all about saving money and not providing any relevant application-specific differentiation, all players will soon be in a race to the bottom: who can do the job for the least money?

Second, if the primary customer motivation is cost reduction, the next logical step would be to ship those virtualized clusters off to some sort of public cloud. Especially if it was super-easy to do so.

Basically, game over for on-premises vendors at that point. Once a workload has gone to the public cloud, there is precisely zero economic opportunity left for any of the familiar hyperconverged players because -- well -- none of them have a public cloud.

This move to the public cloud is not simple hand-waving on my part. It's way easier than you think, and is being widely used today.

I'm going to use Oracle Ravello as just one example as to why I am resolute in my prediction that -- before too long -- the hyperconverged category will collapse into the black hole of commoditization and irrelevance.

And, as a special bonus, I'll arm you with a new buzzword to impress your friends: cloud hypervisor.

Nothing Like Kicking The Hornets Nest

I always find it disheartening when people who might disagree with me decide to discredit me (or question my motives) rather than take on the arguments themselves. In my opinion, ad hominem attacks are tools of the insecure. I try to avoid that kind of behavior at all costs.

Their basic argument is -- since I work for Oracle, how can anyone believe what I say? While it is true that I am employed by Oracle (and it inevitably colors my perspective), my employer does have the luxury of not betting its future on a single industry category, unlike some of the respondents.

I am thus free to speak my mind, and say things that others can't.

Another respondent wondered if I was instructed to write that post by a higher power (untrue), or if it was some sort of retaliation on my part (also untrue). Sorry, I can't provide evidence for either statement, you'll just have to trust me.

Here's my big thought: the first wave of hypervisors changed the way we consumed physical hardware. And now, the next wave of "cloud hypervisors" will change the way we consume hypervisors.

Welcome To Oracle Ravello

Let's say you're running vSphere clusters, or perhaps some of the newer KVM clusters.

Maybe you're out of capacity, out of budget, or both. Cloud, anyone?

You'd like to keep some of your more important workloads around, but you wouldn't be opposed to pushing a few of the less important ones out to a public cloud as long as (a) the process was trivial, (b) the cluster configurations themselves weren't modified in any way, and (c) you had some decent choices as to which cloud and which location.

In a nutshell, that's what the Oracle Ravello cloud service does.

It encapsulates entire application clusters, abstracts compute, network and storage, and then gives you a choice as to which public cloud instance you'd like to run on -- optimized for cost or performance. In addition to Oracle Cloud, AWS and Google are supported.

Oracle Ravello does not require any hypervisor license. It does not modify your VMs or their network/storage configuration. At any time, you can pull your environment back on-premises, or push them to another cloud. It does require that you use their GUI (or APIs)  to monitor and control your VMs.

You pay Ravello for what you use, Ravello pays the public cloud provider. A simple estimator helps you to figure out what it's likely to cost you.  That's about it.

The ideal use case, obviously, would be for people doing development and test on larger, multi-VM application landscapes. And, indeed, that's one of the primary use cases where it is being used today. Software demos and training environments are popular, as well as considerable amount of low-end production workloads.

The service has been available for quite a while, and has an impressive roster of users all with glowing reviews. And anyone can try it for free for 14 days -- no credit card required.

A Bigger Lens?

Students of IT history will recall that hypervisors got their start in development and testing environments.

HVX (the technology behind Oracle Ravello) can best be described as a "cloud hypervisor", as it encapsulates a given virtualized environment and completely abstracts away the architectural differences between various public clouds.  It too is gaining a strong foothold in exactly the same way as legacy hypervisors before it.

As I mentioned above, the first wave of hypervisors changed the way we consumed physical hardware, all for the better. I would argue that this next wave of cloud hypervisors will change the way we consume legacy hypervisors.

We will no longer be bound to a particular flavor of hypervisor (if we choose) much in the way we aren't bound to a particular flavor of x86 hardware today.

Put differently, the on-premises hypervisor wars are coming to a close; as well as the hyperconverged architectures built around them. Case in point: VMware's latest cloud strategy, fresh from VMworld. Personal reaction: quite underwhleming and disappointing as I've always been such a fan.

To me, it was a clear sign of a changing of the guard. The IT world is evolving, and there isn't much VMware can do about it. One of the stories told about VMware's founders is that they wanted to create software that would be used in every data center.

They did that, and were amazingly successful.

But what about a world where people don't want to own data centers any more?

A Hypothetical Rebuttal?

The on-premises hypervisor and hyperconverged vendors will justifiably point to features X, Y and Z that the Oracle Ravello service doesn't suppport today. Gotcha!

My pre-emptive response: (1) not every workload demands those features, so why pay for them? (2) the Ravello roadmap is quite healthy and very agile, and (3) the service is awfully cheap/easy, and works precisely as advertised.

Oracle Ravello may be the first successful example of a cloud hypervisor, but others are sure to come before long.  Welcome to a new industry category.

In a world of cloud hypervisors, it just got all that more difficult for legacy on-premises vendors to justify their existence.

And if your primary reason for existing is cost reduction, best of luck competing with public cloud economics.

 

 

-------------------------------

Like this post?  Why not subscribe via email?

 

Comments

[Michael Endrizzi]

I am a loyal Ravello user at work, AWS certified engineer and user, all my personal apps/data in the cloud, read your blog and love the simple style. However, permit me to argue your point.

1) Microsegmentation - Separating by VLANs and firewalls is dying as fast as data centers are dying. The complexity of coordinating L3 routing with firewall rules and dynamic IP addresses does not scale. Microsegmentation is the next stage. Firewall vendors barely have microsegmentation working on VMware much less the public cloud. AWS only has the equivalent of IPchains which does not scale. It will be a while.

2) VLAN hell. - The limits of using VLANs are bringing it to its knees. 4096 max and limited IP subnet sizes to /23 because of broadcast domains require complex routing configurations the restrict deployment speeds. VXLAN with its suppressed broadcast domains allowing one to build a single 10.0.0.0/8 in theory is L3 Cisco killer...when it works. I don't see that in the public cloud.

3) "My app is slow. May web session randomly works" - I've had too many of these calls trying to figure out who's fault it is. Is Ravello slow? Is Google Slow? Is the Internet slow? I'll call AWS and ask...hmmm no one answers the phone or they laughed at me.

4) Logging - Trying to debug AWS or Ravello via logging is just painful because they control what they log. Try debugging a AWS firewall log.

5) $19K for 10TB - Try storing 10TB into AWS's cheapest storage media Glacier. Pretty cheap huh? Now try pulling it out. Probably about 2 months of your salary. How many of these suprise bills are you ready for. I had a similar experience with Ravello but not nearly as bad just unexpected.

6) Security - AWS has XX number of certifications. Let's hypothetically say our Russian Chinese friends are loose inside of AWS. Do you think you'll get a phone call?

7) Single pane of glass - While you are right data centers are dying there are trillions of dollars of lagacy apps that will never get migrated to public or private cloud. They need to be maintained. How many monitoring apps do you want your help desk to monitor? How many technologies can your help desk manage? Ooops the Python monitoring script failed to report the AWS deployment problem and Johnny GeekHead no longer works here, can 18 year old Fred Jones find it or fix it?

8) Talent - The young crop of techies entering the field are amazing. They program in 10 languages while eating lunch, know firewalls, BGP routing, L2 spanning trees, AD schemas, etc. The people that build and maintain public/private cloud will have to have 10 diverse talents vs the silo'd knowledge stack currently in industry because public/private cloud has centralized control over ALL these technologies. We just need more of these young ones so I can retire and get my hair transplant.

So yes you are right its coming, but public cloud is not a panacea for all apps at this time. Make sure you do your homework before you jump off the cliff.

dreez

Yes

[Chuck Hollis]

Excellent comment, so thanks!