Cisco - you know Cisco?

I was talking to two different friends who both think I'm out of my mind for suggesting that Hostway (as well as every other hosting company) could/should virtualize their data centers so that physical servers can be moved/replaced with no impact on customer uptime. I was reminded that NOBODY we know does such things.

But we know Cisco, and Cisco's a huge proponent of virtualization. I just read Christofer Hoff's post on Cisco developing virtual switches for VMWare's ESX Server virtual network. Other vendors are sure to follow. As Nicholas Carr predicted last year, data center hardware is morphing into software - just as that answering machine you used to have has been replaced with your cellphone provider's voicemail service.

BTW, Christofer wrote an awesome post last month comparing virtualization with walking upright:

1. University of Arizona professor David Raichlen suggests bipedalism evolved because it used less energy. He says humans walking on 2 legs consume only 1/4 of the energy that chimps on 4 legs do. Efficiency gains from walking upright led to exponential capabilities in the long run because early humans were able to use the extra energy to develop better hunting/gathering skills, learn to use tools, etc.

2. Virtualization makes it possible to operate IT infrastructure with less electrical and human energy, allowing more resources to be allocated to  more enabling capabilities. Looking beyond this first step, Christofer sees "grid and utility computing paired with Web2.0/SaaS taking us to a whole different level".

So, back to my skeptical friends who think that dedicated servers work just fine, even if they might be a bit of a hassle to migrate. Just as our early ancestors were reasonably mobile, even if walking on 4 legs took more energy? Google, Microsoft and Amazon are already making their way to that whole different level. Shouldn't you?

Comments

[Jeff Cornejo]

I think that virtualization is a great direction in which to head but the reality is that clients are still not comfortable adopting virtualized environments for their mission critical applications.

Over time clients will begin to see the advantages of virtualization, but the products and market need some maturity.

[Isabel Wang]

A few months ago I moderated a panel where the speakers included a sales exec from Cisco and a professional services exec from HP. I mentioned reading somewhere that 92% of enterprise IT managers are interesting in virtualization, and they both wanted to know where the research firm found the other 8%. They'd worked with many, many organizations and hadn't come across any who weren't looking to adopt virtualized environments.

I think it's hosting providers - not customers - who are virtualization-averse. And I would know. While working at EV1Servers, I decided to discontinue offering Virtuozzo because it was too much of a challenge to support. Our provisioning, IP allocation, license inventory, support ticketing... systems weren't equipped to handle multiple virtual environments on one physical server. Besides, I was swamped with dedicated server sales inquiries. It seemed easier just to stick with what we knew how to do.

But what's easier for hosting companies is not what's best for customers. Check out this disaster recovery case study (it happened in 2004!) on how virtualization helped keep the Las Vegas Water District up and running. You can't work that kind of magic on physical servers!

http://searchcio.techtarget.com/originalContent/0,289142,sid19_gci1232371,00.html

I wholeheartedly disagree on two points.

First, from my point-of-view I disagree that hosting providers are virtualization averse. We offer it up right next to our dedicated hosting and clients chose dedicated hands down, even with the significant cost differential. They don't, yet, understand the benefits. They feel that having their resource usage constrained is somehow a rip-off. All other arguments in favor are null and void. :(

Second, I'm not sure I understand why the hosting provider is at fault for selling the client something they are seeking. They can make the choice, and given all the information still chose dedicated servers. I feel like you are comparing us to grocery stores that sell cigarettes to people even though the boxes have warnings all over them and there is abundant information about the detrimental effects.

[Isabel Wang]

But clients are NOT given "all information". Many hosting companies sell VPSes as an option that's cheaper than dedicated and higher-end than shared. Very, very few articulate the benefits of virtualization.

Back in 2004 I had a conversation with SWSoft CEO Serguei Beloussov. He suggested that instead of private racks, I could offer customers virtual private racks - which would accomplish the same goals of having separate web/app/db servers and interconnecting everything at the backend.

I responded that customers don't want virtual private racks. I had a huge backlog of real private rack orders that I needed to go process.

But did I explain to these customers what their options were? No! And were they mad when their servers died and they had to wait while we set up a replacement? Were they disappointed to hear that there's no auto-magic way to transfer the data on their old Celeron to that shiny new Dual Xeon they just ordered? You bet!

[Christofer Hoff]

Isabel:

I think you nailed the barrier to entry in terms of the "mainstream" adoption of virtualized service delivery for staple functions:

Abstracting service composition, provisioning and deployment are still very difficult for the operational folks to grasp. Transparency and availability (and then confidentiality/integrity) are the focus of these folks; they get paid on uptime. They like to isolate fault domains to singular entities. So, if a single server fails, they know where, what and why. If a virtualized service platform fails, there is the impression they will not be able to isolate the fault and repair it.

The customers really shouldn't care so long as their SLA's are delivered.

In the security space, as an example, virtualizing security when one has to evangelize its benefits is an exercise in repeated head-banging. Some of the reasons are valid, some are absolutely fossilized.

Even if one can illustrate profound CapEx and OpEx reductions with resilient and reliable virtualized infrastructure, the "one goesinta and one goesouta" proponents curl up in a fetal position and draw corner cases that cannot be defended. It's the classical reduction to the rediculous.

This *IS* changing. It *MUST* change.

I spent the last 1.5 years dealing with the office of the CTO on what has to be the biggest virtualized service delivery architecture for IP/IMS converged networking I've ever seen. 18 Billion pounds worth...those who do not make investments (some of them gut-wrenching) to push forward this new architecture will surely wish they had when they are out-gunned and out-outmaneuvered with dynamic on-demand infrastructure.

It's happening. And once it's successful, it *will* take off.

...and as much as I dislike Cisco's marketing lip-service regarding vapor versus deliverable, they ARE making inroads. For that I give them credit.

/Hoff

[Kurt]

Check out Engine Yard; they use virtualization exactly as you suggest. They are a specialized hosting company in that they host exclusively Rails apps, but the founder regularly gives expletive-filled talks about how everyone should be using VMs for all the reasons you list.

I'm not even a customer of theirs, I just think they're smart, competent folks.

[Isabel Wang]

Kurt - I know of Engine Yard, but haven't had an opportunity to attend Erza's expletive filled talks. But I have listened to an expletive-filled Joyent podcast (they're also very much focused on Rails). Now I'm wondering whether there's a correlation between use of RoR and expletives :)

Christofer - I agree. Traditionalists want to isolate fault domains to singular entities. But isn't it better to build a self-healing infrastructure where applications auto-migrate away from failed systems?

And speaking of transparency, the CTO of a 3Tera customer told me a great story about his network architect's recent resignation. During the hand-off meeting, the guy logged into the company's grid and pulled up the AppLogic canvas. The CTO was surprised to discover that it contained perfect documentation of their entire infrastructure. Every switch, every server behind every application, data ingress/egress to from every single NIC. He was in ITIL nirvana without having to go through the hassle of maintaining documentation...

[Christofer Hoff]

"But isn't it better to build a self-healing infrastructure where applications auto-migrate away from failed systems?"

...and security, storage, data, metadata, processes, state, etc...yes!

The biggest hindrance to this is the canvas to allow transparent visualization of the infrastructure in a way that when you push the 'go' button, there is enough interoperable actionable intelligence across platforms that it will work 100% of the time.

That means supporting infrastructure components up and down the stack (load balancers, routers, switches) as well as compute stacks (OS, applications, data) with storage (NAS, SAN, local) and Security (host, firewall, IPS, DLP, etc...)

We're getting closer, but what we need (and the vendors still don't get) is a common set of API's to allow this to happen. We have it in bits, but not holistically.

I'm still optimistic that customers will drive the vendors closer and closer to this and that technologies like SOA will allow us to provide consistent, reliable and secure messaging buses to programmatically communicate using standards.

Onward!

[Tony Lucas]

I think the key point that is mentioned in a number of these posts is that too many hosting companies (ok, just about all of them) still consider virtualization to be the technology enabler to provide cheaper, cut down dedicated servers.

A long time ago (3 1/2 years) we realised that this was not just the case, and using virtualization to enable flexibiity and scalability of resources was a far more powerful proposition (and allows far greater flexibility than traditional solutions).

Very few traditional hosting companies have picked it up so far (I believe ourselves and ServePath/GoGrid are the two leading the way), however the potential customers we are talking to are very enthusiastic about the idea.

I completely agree with Christofer's latest point though. Virtualizing a system is relatively trivial, getting the rest of the network and storage to integrate with it in an appropriate manner has proved quite a challenge.

We are also keen advocates for creating an open standard for interoperability between virtualization providers.

[Isabel Wang]

Exactly. Because you get what virtualization is for, you can present benefits like scalability/flexibility instead of selling it as cheaper dedicated hosting.

I'm actually amazed at how far we've come. I feel like we're (surely, if slowly) heading towards a better-SOA-ed world where applications and data are no longer bolted onto individual devices. So yes, onward! :)