The Cloud Pitch Meets the Web Team: A Sorting Exercise

Google launched something called App Engine last week. Write an application and upload it. Google runs it on Google’s infrastructure, scaling as needed, with no servers to rent or manage. Amazon has been selling computing by the hour (EC2) and storage by the gigabyte (S3) for two years.

The trade press has connected the dots into a thesis: computing is becoming a utility, and ‘the cloud’ is the future of everything. The thesis has begun arriving in our inbox, attached to forwarded articles, wearing the subject line ‘should we be thinking about this?’

Yes. Thinking. This post is the sorting exercise we are running with clients. What is credible for real organizational websites today, what is promising but immature, and what is premature no matter how good the conference keynote was. The categories matter more than this year’s products, because the products will churn and the sorting discipline transfers.

The idea, without the weather metaphor

Strip the branding and the proposition is rental economics applied to infrastructure. Traditionally, capacity is bought in advance. Servers are sized for your busiest imaginable day and sit idle the other 360, plus the racking, patching, and 3 a.m. attention that ownership implies.

The utility model inverts it. Capacity by the hour or gigabyte, provisioned in minutes, scaled up for the spike and down after, with someone else’s staff minding the hardware. For anyone who read our capacity-planning piece in January, the appeal writes itself. The spike problem is at bottom a problem of owning for the peak while paying for the average, and rental economics attack exactly that.

The proposition is real. The maturity is uneven. Hence the shelves.

Shelf one: credible today

Media and asset offloading via S3. This one has graduated from experiment to recommendation in our own practice. Client video, audio, and large downloads move to Amazon’s storage service and get served from there. The per-gigabyte prices embarrass conventional hosting bandwidth.

The service is two years old, the tooling is decent, and the risk profile is friendly. Assets are copies, not systems of record, so the worst case is inconvenience rather than catastrophe. If your application servers are shoveling video, this is the shelf to shop this quarter.

Backup storage belongs here too. Offsite copies of databases and files, cheap and automated. An insurance product priced like one.

Shelf two: promising, watch actively

Application hosting on EC2 or App Engine. The demonstrations are impressive and the gaps are real. EC2 remains officially in beta, without the service-level commitments an institution’s uptime conversation requires. The operational tooling around it is a build-it-yourself affair this year. Monitoring, deployment, and the boring essentials are all on you.

App Engine is a week old and currently Python-only. And here is the strategic catch worth underlining: applications written to its interfaces are written for Google’s environment, with portability an open question.

Our dependency rules applies with full force. Adopt with enthusiasm, architect for divorce, and note that on this shelf the divorce terms are still being drafted. The trajectory, though, is unmistakable. Teams building new applications this year should design as if utility deployment is in their future: separate concerns, avoid hard-wired assumptions about the machine underneath. That costs little and keeps the option open.

Shelf three: premature

Replatforming the institutional website because the future arrived in a keynote. Your CMS-driven site runs fine on well-configured conventional hosting. The caching disciplines we have covered deliver spike resilience at known cost with known tools. The cloud vendors’ current offerings ask you to trade mature operations for early-adopter operations, in exchange for benefits your traffic profile may not need for years.

Migration has real costs: engineering, testing, new operational learning. ‘The trade press is excited’ does not appear on the benefits side of any accounting we would sign. When the shelf-two gaps close, this math changes, and we will say so in this space. That is what watching actively means.

Questions to ask any cloud vendor

Since the pitches will keep coming, the interview questions we would bring to any of them. What is the service commitment, in writing, with remedies? Beta labels and best-effort language are fine for shelf one and disqualifying for anything critical.

Where does our data live, how do we get it out, and in what format? Exit answers reveal more than entry demos ever do. What happens when you have a bad day? Every provider has bad days. The mature ones have postmortems you can read.

What does the bill look like at three times our expected usage? Utility pricing is a feature until a traffic spike makes it a surprise. Model the spike case before it models itself. And the portability question, asked plainly: if we leave in two years, what do we rewrite?

Vendors comfortable with these questions are building for the long relationship. The others are building for the demo.

One relationship note for the transition years ahead. None of this ends the conventional-hosting conversation. It complicates it productively. The hosts that thrive will absorb utility economics into their own offerings. The ones that compete on inertia will discover that inertia is no longer priced where it was. Watch your own host’s roadmap with the same eye you watch the cloud vendors’. Everyone in this market is about to be graded on the same curve.

The pattern to keep

A closing observation from the strategy chair, because we have seen this movie’s structure before. Infrastructure keeps moving down-market. Capabilities that were enterprise-exclusive keep becoming commodities. Analytics did it. Testing tools did it. Computing itself is doing it now.

The winning approach each time has been the same. Adopt the low-risk edges early. Design current work so the middle shelf stays adoptable. And let someone else’s production incidents mature the platforms before your mission-critical systems arrive. First movers pay tuition. Deliberate followers inherit textbooks. For organizations whose websites are infrastructure but not product, which describes many of the clients we serve, deliberate is the right speed.

We will keep the shelves updated as the products churn. The forwarded articles are welcome, incidentally. Sorting them is, increasingly, the job.

One hedge. The shelf assignments above will date quickly, because the market is moving fast. The sorting method is what lasts: risk-tiered adoption, exit paths priced before entry, and enthusiasm calibrated to service commitments rather than keynotes.