There was a question from the audience: “Gee, Steve, what’s the difference between what you’re proposing and straight up virtualization?”
Good question. Glad you asked. Good enough question in fact to insert part 1.5 inbetween parts 1 and 2.
The definition of cloud computing remains nebulous at best. We’re entering a phase where everything is claiming to be a cloud — if you offer something hosted, it’s a cloud. By such a loose definition, the tech biz has been selling clouds since we’ve been renting mainframe time. To offer a little contrast, Amazon EC2 is a huge cluster of virtual machines that you rent a-la cheap dedicated virtual servers.
When you get right down to it, therein lies (in my opinion) the crux of the matter. The difference between cloud computing and JBOViM (Just a Bunch Of Virtual Machines) is a provisioning and resource accounting model. A cloud has granular provisioning with a corresponding accounting model. Those of you familiar with Unix may find parallels with its accounting model which allows administrators control disk, memory, and CPU utilization on a per-user basis with usage accounting embdedded in the kernel.
Not too many years ago, scientists would purchase compute and storage resources from clouds all the time. It was the only practical way for most to run experiments on uber-fast machines. Provisioning and accounting lacked the shiny gloss of EC2′s UI, but the procedure was more or less the same: buy CPU time, buy storage, buy access to tools, libraries, and infrastructure.
In an enterprise, setting up a JBOViM means IT regains the granularity it lost when provisioning servers at a hardware level. What’s next is billing — something I fully expect to emerge from one of the big three virtualization players. Given how quickly a new VM can be brought up and taken down (Xen+NetApp FlexClone is very cool), enabling a business unit to expand and contract its server needs with corresponding billing is a no-brainer that, as I said in part 1, addresses a whole mess of issues in one go. (Security: solved. Privacy: solved. Backup/Restore later: solved. Identity Management: solved. Load balancing: solved. The list goes on…)
From IT’s perspective, the challenge of having to deal with each business unit individually and plan purchasing around it is significantly eased since it is now possible to look at the enterprise’s computing needs as a whole and purchase accordingly without care for specific concerns around CPU and Memory configurations much in the same way that SANs solved disk allocation and sizing issues.
So can an enterprise have an honest to goodness cloud of their own (as opposed to JBOViM)? Yes. And it makes a heck of a lot of sense to do it.
Thanks for the detailed response. For me I take a slighly different spin on it. Bewlow is an except from a recent blog post debating the term “private cloud” .
“It depends if you classify cloud computing as Internet based computing or the Internet as an infrastructure model. I believe cloud computing is a metaphor for the Internet as a infrastructure model, therefore a private cloud is applying that model to your data center, whether it’s closed to the outside world or not is secondary.”
G’day Steve,
Good to see you’re still well ahead of the curve post-Netscaler. I didn’t have time to ping you on my last trip so let me know if you make it to this side of the pond at all.
While it’s refreshing to see someone on the consumer (rather than the vendor) side of the fence preaching about emulating cloud computing internally, I still think it’s a short-sell of the technology. For a start you’re talking about VMs which aren’t granular enough and which carry a significant overhead penalty in terms of the operating system (ala IaaS) and the solution stack (ala PaaS). I prefer the generic term ‘workload’ which could refer to anything from a complete virtual datacenter (ala Qlayer) to a virtual machine (ala EC2) or the application itself (ala AppEngine). As a consumer I need not care what container it runs in, but I may have application dependent requirements (as you do) relating to geography, performance, latency, security, etc. These will impact the cost and given the hypervisor is all but commoditised already will be one of the few areas which differentiate vendors (otherwise it’s a race to the bottom). As was the case when the grid arrived it will be virtually impossible for businesses to internally compete with cheap, reliable utility providers, though some on-premises installations (like the coal power plant at Eastman Kodak) will remain well into the future.
In summary I think the real value will only be unlocked as we move further up the stack. Yes, virtualisation will evolve in the mean time and people will build next generation data centers – it’s even a sensible strategy in terms of easing the transition (P2C and V2C are today’s P2V) – but claiming that it’s a drop-in replacement is a stretch.
Cheers,
Sam