In my blog, “Making Financial Sense of PaaS,” I provided an analysis of delivering a newly-developed mobile application using a variety of platforms. Subsequent to my posting I had some great conversations about the content with Brent Smithurst (@brentsmi) from ActiveState and Mark Thiele (@mthiele10) from Switch. Brent is from a PaaS software provider and Mark is a world-renowned expert on data center operations and architecture, so clearly their insight is extremely relevant and credible.
As pointed out by Mark, it’s very easy to make the cloud look financially attractive when pricing out a single application versus a portfolio of applications. Indeed, I would have to agree one of the most difficult things is to formulate an apples-to-apples comparison of cloud to data center. Even with the concept of reservations, the cloud cannot come close to the amortization of capital allocations across a portfolio of applications. The cloud is all about sizing and costing one application at a time whereas data centers should be all about economies of scale.
For the original blog I chose to use a model where capital allocation was occurring on a project-by-project basis. This is a fair estimating technique given many businesses and government agencies use this technique as a way to procure hardware and software. However, it does skew the results in a particular direction. Additionally, as noted by Brent, the original $25,000 price tag could buy the equivalent of $48,000 worth of IBM BlueMix in GB-hours. Then, finally, Mark noted that even the Hosted PaaS has some operational component to them. He is correct here and that was a clear oversight in the original estimate. IT Operations should be reviewing the logs from the Hosted PaaS instance and monitoring the health of the application and its performance.
So, while one cannot argue the original estimate given the stated assumptions, it is certainly not the whole picture for businesses that leverage virtualization and have pooled hardware resources. Since my labor estimates were based on time, I believe these are firm regardless of the approach taken. The original estimate does not imply that individuals would be hired specifically for this application. Alternatively, if we are going to leverage economies of scale and use common platform requirements across a set of applications, then we have to recognize there will be increased time required of the application infrastructure specialist and operations to handle the increased complexity. That is, as we move away from a single application running in the N-tier environment to multiple applications we need to deal with clustering, high-availability, scaling and capacity management issues that were not accounted for in the original design estimate. This increases the amount of time and, hence, costs associated with these roles.
Taking all these elements into account, the following is an estimate for a business with sunk costs in an existing data center, can add the necessary virtual machines to an existing pool without incurring significant new charges, and will leverage the software licenses across multiple applications. As you can see it changes the financials enough to consider all of them viable options. However, PaaS options could save on average $100,000 per application, which, for IT shops already spending on the order of 90% of their budget to keep the lights on could open up major opportunities for transformational activities.