Clouds, Impressions and Pork Bellies

With Microsoft’s (sort of) biennial PDC on the horizon, my mind (and the minds of many of my colleagues) turns to our cloud computing efforts, which will have their coming-out party in Los Angeles at the end of the month. Like anybody else here at Microsoft, there’s little specific that I can say about these efforts before the conference; all I can say is that we’re working on ways of making it much, much easier to develop,deploy and pay for apps in the cloud.

One of the things I’ve been thinking about, however, in the context of cloud computing, is how it may or will change the way that IT infrastructure (by which I mean, processing power, storage and bandwidth) is bought and paid for. Most cloud or utility-computing offerings in the marketplace today are priced on a consumption basis (that is, you pay for what you use, and no more) – indeed, some people hold that you’re not really doing cloud computing if you’re not charging for it on this basis.

This model for charging represents a significant transfer of risk from the customer to the vendor: whereas an enterprise might today purchase so many servers, and so many OS, database and other software licenses to support a particular service, knowing that some will not be used, now it is up to the cloud vendor to predict demand for their services and purchase the appropriate hardware and software.

But I think that cloud computing may yet enable its customers to further reduce the risk they face, by enabling the trading of futures positions in compute power and storage. And in this respect, cloud computing shares some very interesting characteristics with the online advertising business (Ah, you say, now I understand where he’s going with this). Please note,  by the way, that nothing that follows is intended to indicate any specific Microsoft plan in this area. This is just me riffing.

Clouds as commodities

So, imagine you’re running a news and current affairs website. And further imagine that, oh, there’s an election coming up later in the year which you’re confident will generate a big spike in traffic. If you’re running your site on an on-demand cloud infrastructure, then you’ll be confident that your site will scale elegantly if you get traffic spikes – but at what cost? You may be able to buy compute capacity at (say) $0.10 per processor-hour (or whatever measure of compute capacity emerges) on a spot basis; but if you were to reserve this capacity on a forward basis (i.e. a few months in advance), you could pay only $0.05 per processor-hour.

But what if the spike never materializes? You could just release that capacity back to the cloud vendor and get some number of cents on the dollar for it. But an alternative is that you could in theory sell that pre-reserved capacity to someone whose need is greater than yours, potentially at some profit.

Now consider the same business from an advertising perspective. Anticipating the spike in traffic, you want to sell your anticipated inventory for the best price – which means striking a number of ‘guaranteed’ deals, where you commit to delivering the impressions during the time period (and, given the nature of advertising during an election campaign, you really don’t want to be delivering make-good ads after November 4). So to hedge the risk of not meeting your projected impression goals, you buy a block of inventory that you can use, if necessary, to fulfill your obligations.

As the election looms, however, you discover that your traffic is exceeding your expectations – so you don’t need the inventory hedge. You could choose to take a little revenue from this inventory by serving discretionary ads into it, or you could sell it on to someone else whose inventory prediction was not so on-the-money, and needs inventory to fulfill a guaranteed deal. You could potentially get a better rate doing this than by serving remnant ads into the inventory yourself.

What these two examples have in common is that the publisher is taking a forward position on a commodity in order to mitigate against risk on the supply or demand-side of their business. Of course, this kind of hedging is nothing new – the Chicago Mercantile Exchange has been enabling it for years, for commodities as diverse as pork bellies, oil and coffee. And since energies futures are such an important part of that market, it shouldn’t be a far-fetched idea that compute power (which many have described as moving to be a utility, like electricity) could move to being traded in the same way.

More options

The model even lends itself to the idea of options trading – in both the above examples, the publisher could pay for the option to purchase compute capacity or advertising inventory at a particular price, rather than reserving the capacity or inventory itself; and those options could then be sold on later (or exercised, or left to expire, of course).

The next logical step from there is that folks who have nothing to do with online advertising or cloud computing could start buying and selling these commodities and securities with a view to making a profit on price changes. The economy’s current woes notwithstanding, I can see this happening in the next 5 – 10 years.

To make either scenario a reality, however, there need to be functioning exchanges for the buying and selling of the commodities. This is close to becoming a reality for online ad inventory – the likes of Right Media Exchange, DoubleClick Exchange and our own AdECN are close to providing open trading platforms for advertisers, publishers and networks to buy and sell inventory. Though there is no talk of futures trading in these environments right now.

It’s significantly further off for cloud computing capacity. For a start, the industry lacks standards for measurement and billing – will it be the processor-hour, or the Gbyte-day, or the Gbit-month, or some combination of the above? Secondly, unlike the online ad market, where a given ad will run on most publisher sites (with the exception of rich media ads), there is illiquidity between different technology platforms in cloud computing – so an app written for Amazon Web Services will not run unmodified on Salesforce.com’s cloud platform, or Google’s. This may never change, in which case any kind of market or exchange for compute capacity will be limited to a single vendor’s system, greatly limiting the effectiveness of such an approach. But interesting to think about, nonetheless.