Virtualization Automation – A Proper Virtual World

When Is It Time to Say Good Bye?

October 8th, 2009

So The Old Virtualization Platform Is Just Not Doing It for You Still (Part 1 of 2)

This time of year a lot of hard decisions are made in enterprise firms, I am sure many will identify with this statement. The New Year is approaching, the budget is due, the bonus if any is on the line, and now is the time to pull that rabbit out of the hat. There are a number of reasons to keep or leave your past virtualization platform, this article will explore a few points of on why a platform should or should not be continued, with the premise that it is time to say goodbye. The basic logic is evaluative analysis, something any of us in the computing industry know or should know well. The two key questions are:

• Is the solution effective?
• Is the solution efficient?

The basic objective is to save expense, total expense; this includes factors beyond the virtualization platform, and is reflective of the above questions of effectiveness and efficiency:

• Can the platform be revised, improved, or otherwise developed?
• What does management want, really want?
• What is the complexity of the effort?
• What is the time line and scope of the effort?
• Can the customer base survive the transition?

The answers to these questions will be environment specific, again for the sake of illustrative explanation, each question will be explored in conceptual context only.

Is the solution effective? If the solution has been around for a few years, the expected answer would be yes. However, early in virtualization experimentation by a number of firms, mistakes were made, and true cost avoidance or savings may not have been as expected or hoped. Often virtualization platforms are kept conservative, and structured with caution, but over time, expectations for continual economies of scale demand more for less from the original design cloud the results if not the perception of effectiveness.

Is the solution efficient? Just because initial cost avoidance was good, or better than expected, does not mean the virtualization platform was well managed over time. This is an unfortunate, but a common problem for many firms that have significant virtualization. Staff changes or organizational reorganizations often break efficiencies gained as the virtualization platform matured, and just when the solution should be a smooth operating entity, things go wrong. Both management and engineers just cannot seem to leave something that works alone.

Can the platform be revised, improved, or otherwise developed? This should be an obvious yes. If not, then something is crippling the solution, or someone has made some horrible decisions about the virtualization solution. Management is often impatient for results; a short-sighted mind set for virtualization is often a key factor in failure of the solution or lack rational expectations. It takes years for a virtualization platform to mature regardless of the vendor. The vendor needs to be stable, consistent, and responsive, if not, walking away from the solution is more than possible. A struggling vendor is a key indicator that years later nothing but pain will result. The solution should work at reasonable scale, maturing feature set should never mask foundational or core functional issues. Fixing bugs and improving stability cannot be given a back seat to new feature development.

What does management want, really want? Never overlook the fact, that management may just not like the solution in place, the reasons for this are endless, but they often come down to two factors, or issues. The first issue is cost, management hates paying for anything twice, and so what was acceptable at the initial deployment of virtualization, is often a problem a few years later. VMware has this issue today, VMware costs go up, while customers are expecting costs to go down. Feature set grows, but most customers do not use all features or worse are forced to purchase many features just to get a few key needs addressed. This scenario just ticks management off, and is often the reason that a given solution is kicked to the curb even when successful. The second issue is competition, the best solution in the world, and I have discussed this issue in the past, often does not gain the greatest market share over time, Sony BetaMax versus VHS, Apple Macintosh versus PC clones, etc. True, the many will point out the iPod, but in fact, the iPod dominates only because it has no real or true competition, once there is something that is better, or even close to the iPod? Guess what! For example, the iPhone is already facing this issue with the latest generation of cell phones from other vendors, that are approaching iPhone class of service and range of use features.

What is the complexity of the effort? Or in other words, is it easy or hard to walk away from the existing virtualization platform? Today, moving from VMware VI3, e.g. VirtualCenter and ESX, to say KVM or Xen is not painless, but less painful every day. KVM and Virt-Manager make it easier to leave VMware vSphere now that KVM/Virt-Manager support OVF via Virt-Inst, and Open OVF expanding to support Hyper-V. In fact, once OVF supports references to virtual machine disks and does not embed virtual machine disk data, the transportability of virtual instances will be almost seamless, really closer to a classic cold migration, since the common shared storage is leveraged.

What is the time line and scope of the effort? Well, this question is not fair; it is a trick question of course. If the complexity of the effort is near painless and/or management has already decided it is time to go, then does the time and scope of the migration to a new solution really matter? Yes and no. Yes, in the sense that a complex migration may delay the migration. No, in that once the action plan is defined, and executed, the writing is on the wall. It is interesting that Microsoft has not been splashing all over the world, how they are stealing hand-over-fist customers from VMware? Why? Well Hyper-V until 2008 R2 with CSV (Cluster Shared Volumes), was not a true challenge for VMware VI3, never mind vSphere. Of course, KVM is still maturing. Oracle is stuck trying to figure out what to do with SPARC, so Zones never made it to the sand box? But once the pain of migration is resolved, watch out! Microsoft and others learned a number of things from killing off Novell. Microsoft Windows NT server even with faults was a general application server that was easy to management and use, oh and it did file serving as well. Microsoft made sure moving off Novell was a painless as possible, CSNW (Client Services for Netware) and GSNW (Gateway Services for Netware) for example. Novell 4 was a superior solution to Windows NT in serving files and using LDAP, but still it was painful to go from Windows to Novell, compared to Novell to Windows.

Can the customer base survive the transition? Aw, shucks, the end-users, who cares about them anyways? This is the most obvious sleeper issue. What? Wait! What the heck is an obvious sleeper issue? I gave away this one in the answer above… CSNW for example, let a desktop think it was working with a Novell server, when it was really a Microsoft Windows NT server. It was about as transparent and seamless as possible considering the radical differences between Novell and Windows NT. CSNW make the end-user experience, if everything was done right, painless as possible. Now consider virtual instances, which are several steps away from hardware, and even platform specifics, combined with V2V (Virtual-to-Virtual) tools and methods? Or even easier, a simple reboot, because KVM can support VMDK files? Sure KVM VMDK support many not be the best performance, but on the standard pain scale, is more like of an itch to be scratched, than a needle prick. OVF is another option as well. Unless someone is just plain goofy, customer impact should not be a factor.

As for the key question of this article… When Is It Time To Say Goodbye? When the answer to all of the questions above are… No big deal. Sure, many will say that the management demand issue trumps most if not all? Well, I disagree. Virtualization platforms that are run well, work well, and avoid cost, will and do take years to implement, taking years to retire, at an enterprise scale deployment with 100s if not 1000s of hypervisor hosts, and 10,000s if not 100,000s of virtual instances. Moreover, there is one superordinate concept that will take most if not all of the remaining pain out of an enterprise scale migration to different virtualization platforms on demand, wait for it, wait for it… Stateless! The emerging acceptance and implementation of stateless computing concepts, at a hypervisor level as well as at the guess OS level in virtual instances, is or was the last technical foundational stone holding virtualization platform mass migrations from taking place with little or no pain. Is 2010 going to be insane or what? Will it be the year of mass virtual migrations? I think so? Do you?

Entry Filed under: A Proper Virtual World