For a few weeks now, I've been offering some thoughts on public cloud pricing, going so far as to say that infrastructure-as-a-service is a bad deal--which, over time, and sold the way it is now, it is. But that doesn't mean IaaS is never useful. It does have good uses, and I'll describe some here, but it's also likely to evolve into something more interesting.

Hewlett-Packard last week began to describe the services it will offer in its own cloud, beyond the standard fare from Amazon Web Services (AWS). Some of those HP services sound good; some sound problematic. But before I dive into those, here's a quick reminder of the problem with IaaS.

Buyers of cloud-based storage and compute services are paying for something whose costs become cheaper on a logarithmic scale over time, but cloud provider prices now decline only linearly over that same time. Under those conditions, sooner or later, it becomes cheaper to own and manage the resources yourself than to buy them from a third party.

Recent evidence shows that big customers are starting to reach the same conclusion. In his InformationWeek report on cloud ROI, contributor Jonathan Feldman points out that online game maker Zynga recently moved from an 80/20 mix of cloud/in-house data center resources to a 20/80 one. Similarly, DreamWorks Studios bought from AWS about 20% of the CPU hours it needed to render animated movie Kung Fu Panda 2, but it did the rest in-house.

[ Learn more about the cloud's value proposition. See Does The Cloud Keep Pace With Moore's Law? ]

If you use IaaS like Zynga and DreamWorks now do, it can be a good deal. Zynga now uses the cloud for proof of concept. It launches a game there, learns what the resource demands will be, and then brings it back in-house when it understands the need. DreamWorks used AWS for bursting (Panda bursting?) in its production work. One can imagine that as a film comes toward deadlines, the cost of using outside resources is fairly easily justified, and there are, no doubt, parts of the rendering process that use gobs of compute power.

Feldman offered an analogy most homeowners will appreciate. He looks at IaaS in the same way he does the local United Rentals store. If you're a weekend warrior and you need a tile saw, jack hammer, or concrete chainsaw for a small project, you go rent one, even though you know that a single weekend's rental price is probably a quarter of the price to buy the tool. But if you lay tile for a living, you buy your own tile saw. The point is that for very infrequent use, the price of the rented tool almost doesn't matter. You need it when you need it, and you don't want to own it under almost any circumstance. If you have computing needs like that, the cloud is for you.

Zynga's cloud strategy removes much of the risk of launching a game because the company can delay the data center capital outlay. Developers can make changes and operators can determine just what infrastructure will be needed when the company decides to bring the game in-house. And if the game flops, it flops. So when Zynga does make a capital outlay in support of a new product, it already knows a lot about a game's chances for success.

As HP describes its new cloud offerings, it seems to implicitly recognize these sorts of uses. In discussing the cloud tools it's offering to support developers, HP mentions languages like Ruby, Java, and PHP. Supporting developers in their use of these languages is an implicit recognition of the types of cloud uses Zynga and Dreamworks represent – that is, for proof of concept, for testing, and for exceptional needs, all instances where developers will be key players. On the flip side, HP says it will offer structured and unstructured database services, as well as analytics.

This last area may be a tougher sell. If HP offers a "big data analysis" cloud, the gotcha will almost certainly be in the cost of storage. Because unlike processing power, which you can more or less use like electricity (on when you want it, off when you don't), storage is different. At OC-3 network speeds, you can move only a gigabyte per minute or a bit more than a terabyte per day. There's no concept of moving data into the cloud "just in time" to do big data calculations; it takes days to weeks over most wide area networks. Big data usually accumulates and stays put, and that puts us right back into the original cloud pricing problem I've been talking about.

The cloud will find its niches, but it will be a very, very long time before "everything" runs in the public cloud.

As enterprises ramp up cloud adoption, service-level agreements play a major role in ensuring quality enterprise application performance. Follow our four-step process to ensure providers live up to their end of the deal. It's all in our Cloud SLA report. (Free registration required.)