The results are in and it seems that our readers are almostleaps and bounds or at a slow steady pace. Unsurprisingly, I received far more comments in direct emails than were posted on the discussion, and given that I took the approach that overhaul was the better approach, the majority agreed with me.
The naysayer's generally brought up two arguments: Cost and risk. The first is hard to argue with; if you can't afford to make major changes, then you are unlikely to consider that as an option. But no one told me that they actually took a step back and considered the cost in terms of the overall expenditure over time and the potential cost savings. And look at what gets done for very little extra expenditure; perhaps your next round of updates could be consolidated to a single aisle, rather than scattered throughout the datacenter.
And on that aisle, you could try one of the hot-aisle containment tricks, using inexpensive plastic sheeting and PVC pipes for a frame. Pick a single business process to dedicate that new setup and run it in parallel with the existing system until you're comfortable with its function. Then sit back and compare the efficiencies of that setup to how you were previously running. Getting it working for you may not be as simple as swapping in new blades in servers throughout the datacenter, but it might just get you on the path to a more efficient datacenter overall, with minimal risk and expense.
It also looks like many folks consider this an "either or" exercise, when it clearly isn't. While it might have been framed that way, no one is advocating the wholesale replacement of systems in your existing datacenter; a clean sheet approach is only appropriate when you are creating a new facility (either in an existing space or brand new). But if you aren't taking modular and what is effectively large-scale appliance computing seriously, your business is going to end up way behind the curve.
"We don't do it that way" because it embodies too much risk is the response from someone who hasn't actually looked at all of the issues involved.
The practicality of a major update to your datacenter operations by adding a turnkey system like one of the Microsoft Parallel Data Warehouse appliances or a Hadoop-in-the-box solution can't be dismissed because it represents a major expense or change in the way you do business. If it is the solution that is going to get you the best possible ROI, based on speed in getting the capabilities to your users, improved performance, and better efficiency for the affected applications, it should be considered.
The same applies to modular additions to your datacenter. Perhaps it is time to look at containerized or modular update, allowing you to make significant improvements with having a negative impact on the existing infrastructure.
"We don't do it that way" because it embodies too much risk is the response from someone who hasn't actually looked at all of the issues involved. And as far as it being a alternative with too much upfront expense, in many businesses, it's little more than the money that will be spent eventually, over time, solving the same set of problems. And how valuable is it to your business to have those problems solved now, rather than over the course of the next year?
In any case, only you and your team know your business well enough to make these types of choices. But don't let your embedded business practices have a negative effect on the way you update your datacenter. When considering the cost, make sure that business management understands the overall impact of major improvements; not just the upfront expenditure.