Thursday, February 12, 2009

Grumpy? Just a layer away ...

It’s difficult to break the old habits!

For as long as I can remember, I have enjoyed an early morning coffee. When at Tandem, I used to routinely walk to the Roasted Coffee Bean when it was still in Cupertino Village.

These days, I enjoy a much shorter walk to the local Starbucks - but the pattern remains firmly entrenched. But lately, here in Southern California, it's turned cold and I have taken to throwing on an extra layer; the picture I have included here was taken of me earlier today wearing a “hoodie” that, should you be able to make it out, simply says “you say Grumpy like it’s a bad thing!”

If I don’t get that first cup of coffee, it can be an accurate reflection of my mood!

When I was on the board of the IBM SHARE user group, we held a 2008 event in Orlando, Florida, around this time of year. It was at the Disney complex and when I packed for the event, it never occurred to me that Florida could be cold. But the temperatures hovered in the low 50s Fahrenheit, and the hoodie remains a legacy from that week. As I was reading my mail, I was reminded of this as I came across a SHARE “5 Minute Briefing” email.

As part of the SHARE President’s Corner there was an article on virtualization, with a link to a report by the Vice President about a press tour in support of the same theme:

“The irony is that the concept of virtualization isn't new. It has been around nearly as long as the modern computer. The most common implementation of virtualization, the hypervisor, has its roots in the early days of mainframes, when computers were big, expensive and scarce,” began the article in the SHARE President’s Corner. The writer then adds “it was developed to help solve the problem of too much demand for computer services, not enough hardware to go around … now the IT industry is turning to virtualization to solve the problem of too much demand, too much hardware.”

If I have this right, when there simply wasn’t enough hardware in the computer room, IT turned to virtualization. As hardware prices continue to decline to commodity price points, with much of the decision-making in the hands of the business units, now there is too much hardware! But again, IT is turning to virtualization. I need another cup of coffee!

Regular readers of this blog will know of my appreciation of architectures that separate an application’s business logic from the underlying technology – and deploying any level of abstraction can go a long way in isolating the peculiarities of a platform. There are trade-offs, of course, with the most significant being the impact on performance. But the inherent flexibility that arises often proves extremely tantalizing to any architect looking to protect the business logic from changes to the hardware, the operating system, and even the data base. Virtualization is one such abstraction layer.

Following the link to the report by the SHARE Vice President, covering the press tour in support of virtualization, I read how “members of the press were interested in trends associated with the emerging use of cloud computing …” Journalists have bad habits as well, and I observed this myself on press tours I have done. From virtualization to cloud computing – with barely a blink, I suspect. But are the two related?

Is cloud computing the next step after virtualization? Is cloud computing better served with the computing environment, virtualized, inside the cloud - can cloud computing even be manageable without virtualization? And perhaps most important of all, can one of the guest operating systems be NonStop – in other words, will there be a role for NonStop within the cloud?

Russell Daniels, HP's CTO - Cloud Services, has been championing cloud computing for more than a year and has been making it very clear to audiences that "a cloud is a network centric application delivery platform, and should not be confused with messages on virtualization; it can be delivered using a virtualized infrastructure, but it's not virtualization." It’s about the coming together of two technologies rather than one developing from the other; it’s about cloud computing leveraging what virtualization can provide – and I firmly expect to see aggressive NonStop participation inside the cloud as cloud computing matures.

I was walking through the mall recently and came across a sign that said “introducing our new buckle to buckle protection plan” and I have included a picture of it here. The sign reminded me of how protection from glossy “Gucci Marketing” efforts, and from jumping from one “high powered marketing” fad to the next, is hard to find. So often we fall for the same old lines, just out of habit!

And reaching for another cup of coffee, I wondered whether I had provided any “buckle to buckle protection” in this blog!

A year ago (February 12, ’08) I wrote the blog posting "My Wish" for NS Blades where I suggested that “my second wish is to see a hypervisor introduced where NonStop can be configured as a ‘guest OS’ in much the same way z/VM is used on the IBM mainframe.” I was suggesting, at that time, that perhaps it would be a good idea to support NonStop running directly on top of a hypervisor – and a clear shift from a previous position.

In an even earlier posting (September 13, ’07) A Taste for Virtualization I had been adventurous enough to suggest that “will this be a big step forward for NonStop? Will it even look like the NonStop product we are all so familiar with – or will the NSK OS morph and become some derivate of the traditional NonStop Messaging System (a la ServerWare, we saw all so briefly back in the mid ‘90s) layered on top of Linux on top of a highly tailored HP-specific hypervisor?”

However, I began to reconsider these earlier remarks and in a more recent posting (September 2, ’08) Blood and corpses everywhere! Really? I reported how “‘NonStop is already fully virtualized,’ one senior manager pointed out to me recently, ‘from the bottom up.’ You are shielded from what’s actually going on at the metal and OS interaction level … ‘putting a virtual machine (VM) under a Massively Parallel Processor (MPP) machine (as we have with NonStop), is actually not helpful. You start creating havoc with the definition of fault zones tied to a logical processor.’”

From NonStop being dismantled, to where it might be a good idea to run NonStop as a guest operating system, to where nothing at all needs to be done (NonStop is fully virtualized), suggests some vacillation on my part. Virtualization has remained a “hot topic” and it is an addictive theme for any hardware vendor with multiple operating system options. “Multiple operating systems? Not a problem - you can run any of them, or all of them, on top of our new Cyborx Hypervisor,” I can hear the pitch. Whatever! So, what happened, and why the apparent changes?

The arrival of Blades broke the habit. And rapidly! As with the emergence of any disruptive technology, looking at what Blades can provide gave me cause to revisit these earlier predictions. And the more familiar I have become with what HP is doing with Blades, the more I realize that NonStop can operate outside of native, or bare-metal, hypervisors.

The way NonStop scales (up or down), processors can be added or removed with no impact on the applications – NonStop shields the applications from any association with the bare-metal. In this respect, it provides a similar abstraction layer as I would expect from a hypervisor. And so, a BladeSystem with 16 Blades assigned to a hypervisor that, in turn, was supporting any mix of Linux, Windows, and Unix, and 8 Blades assigned to NonStop would only need a front-ending workload balancing product to monitor operating system requirements to reassign Blades between the two supporting environments.

In that February 12, ’08 post I asked whether we could see “a variation on today’s workload balancing products, but supporting a transaction profiling capability that, once set up, learns about the overall mix of transactions, and automatically adjusts the OS configurations on the fly?” I then observed how we would “no longer … agonize over the ratio of processors assigned to any of NonStop, Linux, or whatever – the system would learn enough over time and adjust the composition accordingly!”

In focusing on virtualization, the SHARE user group is recognizing the importance of the subject within IT – and it is correct in pointing out how the earliest examples were found on mainframes. But the reality is that IT is facing an explosion in hardware deployment and the solution will require virtualization. None of this is lost on HP either, and as the BladeSystem continues to develop traction in the marketplace, I can only see virtualization being increasingly relied upon for better utilization of the Blades.

NonStop, being fully virtualized, operating alongside of any commercial hypervisor in the same BladeSystem and balancing the usage of Blades based upon demand, looks completely viable to me and with plenty of advantages. After all there will always be mission critical applications and a well-packaged solution that includes NonStop could see new markets develop and renewed growth from NonStop usage.

Hypervisors, as a layer of abstraction, have their limitations – and it’s not only about performance. With NonStop, it’s all about availability – and potentially degrading the bullet-proof levels of availability provided today by NonStop, is one trade-off that needs to be carefully assessed. Unable to support traditional levels of uptime would certainly sour many an IT director, and have them lose interest in virtualization – and with all the other concerns we face today, I sure wouldn’t be happy to see a run on Grumpy’s hoodies.

Footnote: And for my good friend Bill Leistner- there’s another story about Grumpy that I will leave for another time ….

No comments: