As a finance guy, I subscribe to the old adage "There's no such thing as a free lunch." So when even McDonald's has eliminated its "supersize" menu items, it probably behoves us to look with some scepticism at the premise of a supposed banquet in the form of on-demand computing.
The theory of dynamically assigning computing resources across entire environments and applications is absolutely the right direction toward which we should strive. But when the majority of our fellow professionals are literally working overtime addressing simple patching issues, it seems well short of reality.
So putting on my IT guy hat, let's look under the covers.
One of the general tenets that has been consistently validated in consolidating applications to a single physical environment is that a key planning task is to conduct a thorough assessment of the compatibility of the various applications. In practice, a critical success factor for server consolidation is being able to ensure a balance between raw CPU utilisation, I/O and other resource allocations.
The various applications are then allocated resources based on a static assignment recognising the best compromise - which we sometimes, unfortunately, fool ourselves into thinking is optimal, as opposed to simply workable.
Almost by definition, dynamically assigning resources requires a detailed understanding of the modelling criteria under very complex, real-time computing loads. Unfortunately, given that the relationship between CPU and I/O is rarely linear for batch processes, and even less predictable for online transactions, the complexity of maintaining optimality in real time quickly becomes evident. In other words, this is just really, really complicated and not yet ready for prime time.
So much for theory - how does this play out in the real world? Given that the systems can't dynamically reallocate without intervention, the question then becomes whether the savings from consolidation outweigh the increased risk of manually resetting allocations. Like most things in life, the best approach is moderation. For example, consolidating file servers that can be set almost permanently is likely a very good thing - with controllable risk - but mission-critical systems with fluctuating loads is something else.
With modelling tools and true dynamic resource assignment conservatively four or five years out (according to my hardware vendors), on-demand in its current form is much more akin to an electricity monopoly with unabated pricing power and no oversight. The most common formats revolve around installing estimated peak required computing capacity with a contracted baseline and tiered payments applicable at higher use levels (remember utility computing from a few years ago?).
In conjunction with the hardware, there is usually a large component of services -- often consulting services to work out some pattern to reconfigure systems on some predicted basis and some services to reset the configurations (for example, weekly or at month's end).
Given my experience that no user has ever, possibly in the entire history of IT, complained of too large a system, all uncontrolled available resources will be consumed, matched by the commensurate peak prices for the computing power. Alternately, if peaks are managed to remain below the installed maximum capacity, the vendor will need to renegotiate the baseline. As a simple matter of economics, if a vendor installs lots of hardware, the costs will be covered one way or another.
The analogy with a utility holds from the perspective that making sure computing is available in the worst-case scenario is a lot like building massive generating capacity for absolute peak loads. Perhaps, as IT practitioners, we should also consider the utility practice of negotiating or reserving the right to disconnect some customers when required to preserve the network? At least that way we'd have a clear price/value indicator directly from our customers.
So, what's my take-away from all this? When I look at the offer of a free lunch via on-demand computing, I tend to suffer mild indigestion.
Bean is vice president of corporate systems at Hilton Hotels in the US.
Find your next job with techworld jobs