The Cloud OS: PR fluff or tech awesomeness?

The first hostage was the cloud itself

Re: don't allow your company to become a hostage

What the industry that has fought us every step of the way is trying to do is equivalent to designing proprietary networking protocols and locking us into them.

The 'cloud' goes back to the old concept of the X.25 cloud and what we meant when we drew a literal cloud between one system and another was that the cloud abstracted the connection so you neither knew nor cared where it was or how the data got back and forth. The details of transport and location were not relevant to the conversation.

As time went on, those of us making network drawings started to refer to the side of the cloud facing the client as if the cloud itself was an entity that contained the resources it delivered. It simplified the representation of designs to show only what the client saw as services provided by the cloud.

At this point in time, marketing people have begun to speak of the cloud as a thing unto itself -- as if the resources it delivers *are* the cloud. You have to 'get some cloud' now. Some who, like me, have been at this since the 1980s and earlier are starting to get a little antsy about this mutation of the concept. We have been 'moving' to the cloud forever, but it was not a marketing idea.

The movement 'to the cloud' has been underway for a long time. It is an inevitable logical next step as the global network acts less like a long haul communications line and more like a bus.

Companies like Microsoft are fighting like mad to make sure that you don't use the cloud to escape their proprietary infrastructure. Part of their strategy is to confuse the conversation. If people understood this better they would be insisting on vanilla standardized non-proprietary protocols to abstract Storage and CPU cycles so that it was easy to rent portions of their infrastructure as they needed them, without worrying about how that infrastructure was attached.

Ethernet, TCP/IP, and various RFC specs are what have created this rich ecosystem that makes talk about 'the cloud' as a provider of resources make sense. The moment someone starts talking about their proprietary cloud infrastructure you should be wary. Replacing X.25 with TCP/IP was the right thing to do. Enhancing TCP/IP to increase addressing ability makes sense. Replacing or encumbering TCP/IP with proprietary extensions and enforcing what can deliver services, not so much...

Cloud OS

I would suggest there are two broad types of Cloud OS.

The first I think Dale Vile alludes to, namely the real-time, loosely-coupled, highly distributed, highly scalable OS that stitches all the compute nodes together. Any one who worked with Transputers and other massively parallel and/or distributed architectures would be familar with this style of OS kernel. By it's nature it sits below the hypervisor, although todate it has probably been bundled up with the hpyervisor.

The second, I think Cliff Evans alludes to, namely the OS that handles the emergent properties of the Cloud. Obviously to date much activity in this space has been devoted to cloud administration and management, but we can expect this to evolve to include services more directly relevant to applications and data.

PR fluff

At least as far as this article is concerned, two of the three folks participating are pitching Microsoft cloud. (I realize Trevor is independent since I have read some of his other posts but it should be more balanced).

Which is one reason I decided not to take El reg up on their offer to write articles here. I can have my own biases in my comments and blogs, but I feel a higher level of responsibility(whether it is expected or not) when participating in a more formal news article.

As it stands today "The Cloud OS" is pure PR fluff. It has been since VMware coined vSphere 4.0 back in 2009? 2008? I forget the first "Cloud OS" (predating Oracle's claim by years). Maybe there were others before it.

Making a cloud is possible, but I firmly believe most organizations can and do get fine with a slightly older paradigm(never thought i'd use that word). One that has gone out of style, one that really was never in style the way the cloud is - Utility computing. Where you can leverage the capacity of servers, storage and network to drive utilization up(thus significantly reducing real costs on said tier 1 infrastructure) and maintain high levels of availability (vmotion, storage vmotion (or their equivalents) - MAC address takeover for fast network fail over, etc). FAR less complexity -- a little more manual effort around the edges, and you retain the ability to maintain control - often times cloud systems remove capabilities in order to compromise in other areas.

My boss made an interesting comment to a group yesterday, he said we are working to double our VM capacity in our environment and we are doing this at a cost of 15% of the original purchase(never really thought of it in that angle), rather than doubling the cost. We aren't adding a single server to double this capacity either (thus no additional vSphere licenses).

Cloud Latency...?

"The concept of the cloud OS lashes together various technologies under a single banner in an attempt to drive people towards a subscription-based consumption model for their computing needs."

Great. But why does no one delve into latency? You've outsourced your IT and saved on costs. But what about the wider implications? I've asked before: It sure is an attractive idea to outsource menial daily maintenance and overall worry to Cloud providers. But I rarely glimpse any insight into the impact of latency? Can someone offer some real-world enlightenment? Do all companies take at least one critical app, port it to the cloud, and then do benchmarking before migrating?