12 Comments

Summary:

Cloud services share similarities with other on-demand, pay-per-use offerings such as airlines or car rentals. But what’s past in those industries may be prologue for the cloud. Here are some key aspects of those services that could become integral to the cloud.

With the second decade of the millennium now just weeks away, I thought I’d offer up some possibilities for the cloud computing market as it continues to evolve. Cloud services — whether infrastructure, platform or software — share similarities with other on-demand, pay-per-use offerings such as airlines or car rentals. But what’s past in those industries may be prologue for the cloud. Here are some key aspects of those services that could become integral to the cloud in the coming decade:

Non-uniform Pricing — Since the costs of real estate, power, cooling, carbon emissions and bandwidth may be location-dependent, shouldn’t prices of cloud services vary based on cost differences between locations? In addition, a hotel room with an ocean view is priced higher than the one across the hallway next to a parking lot, and not because the mattress costs more, but due to value-based pricing. Similarly, location can mean everything when latency is important, which is why some cloud providers are offering services near stock exchanges where microseconds might mean millions. Time-based pricing should also come into play. Shouldn’t computing cycles at 2 a.m be priced lower than ones at 2 p.m.?

Volume Discounts — Buying more resources at a given time, or the same quantity for a longer period, should entitle the customer to a lower price, since the risk of unused provider capacity is lower.

Reservation Protocols — Customers who show up at a hotel without a reservation risk sleeping in their car. Hotels that accept reservations with no or refundable deposits risk no-shows and lower utilization. Reciprocal commitment exacted via retainers, pre-payment, or non-refundable guaranteed reservations enhances provider financials and provides benefits to customers such as assured availability and discounts.

Oversubscription — Buying an airline ticket is different than flying. If not all reservations are used, providers can maximize revenue yield by overbooking, even if credits or penalties are occasionally paid. And, this is not a bad thing, since real users of limited resources are not blocked by users with unactualized intent.

Space-Available Upgrades — Airlines award empty business class seats to frequent flyers, enhancing customer loyalty. Perhaps cloud providers should consider the same, e.g., a free extra copy of a data object, space permitting?

Dynamic Pricing — Finite, perishable capacity — such as airline seats and hotel rooms — drives firms to use sophisticated yield management algorithms to maximize revenue, reducing prices to increase demand when utilization is low, and raising prices when utilization is high. Congestion pricing to discourage peak use — whether of city streets or electricity — and promotions, e.g., sales, to encourage use help smooth demand, improve utilization, and therefore optimize economics.

Capacity and Rate Transparency — Dynamic pricing requires rate transparency. No one wants to be surprised when the bill comes. “Click to view available seats” and “five seats left at this price” provide customers information that can help them plan, or accelerate, purchase decisions.

Discretionary Processing and Auctions — Much computing must be done on demand. However, in the same way that $79 fares to the Caribbean make people reconsider their weekend plans to shovel snow, companies increasingly will be able to decide how much processing to do for some workloads by placing auction bids or based on spot prices for computing — consider complex optimization problems where more computing results in better results, but with diminishing returns.

Derivatives and Hedging — Options and futures exist for equities, commodities and currencies ranging from pork bellies to pesos, so why not derivatives for network, compute and storage? Jet fuel is strategic to airlines just as IT is for many firms, and the same way airlines hedge against jet fuel price increases, an e-tailer might hedge against CPU core price increases for the holiday season. Such futures and “options for the cloud” could mitigate price risk, via hedges that protect against dynamic pricing and market vagaries.

Markets — How to trade capacity and derivatives? Why, spot markets and auctions and option markets, of course, such as BuySellBandwidth.com. Cloud service providers or enterprises may want to trade future capacity and protect against smoking hole disasters by acquiring options for capacity from other providers. Coming soon, the New York Server-Hour Exchange?

Volatility — As recent times no doubt illustrate, meltdowns and irrational exuberance are inherent to markets composed of traders with attitudes. In Ubiquity, Mark Buchanan reports on research conducted by two “econo-physicists,” who modeled a simple market in one stock, with three types of traders — optimists, pessimists, and value investors — who could shift their orientation. Even in such a simple model, bull, bear, and chaotically volatile market behavior emerged.

Aggregation, Cooperation, Brokerage, Arbitrage, VARs, and Other Intermediaries — New market ecosystem roles will evolve. Aggregators may buy capacity at volume discounts and resell smaller quantities (“break bulk”) to make money. Cooperatives such as the Enterprise Cloud Buyers Council may wield buying power to save money. Like travel web sites, brokers will arise to resell, package, compare, crowdsource reviews, and recommend providers, and VARS will add value to wholesale cloud computing capacity. Arbitrageurs and “high-frequency” traders may arise to make money on instantaneous market imbalances.

Virtual Cloud Operators — Airlines sell codeshare partners’ capacity, increasing the apparent breadth of their portfolio and boosting revenue. One model has SaaS (software as a service) providers white-labeling other SaaS providers’ offerings. Or SaaS might run on another provider’s infrastructure as a service (IaaS) — virtual IaaS operators might physically reside on other operators’ infrastructure.

Co-generation — If major companies can generate power for the grid from their generators during lulls in internal demand, they may also be able to sell their unused compute capacity, as GridEcon proposes.

Simplified Pricing — There is a natural cycle at work in market ecosystems, where fixed pricing drives the introduction of pay-per-use; pay-per-use becomes increasingly complex; and this may drive a return to simplified plans. In telephony, pricing started out as fixed (per-line), then became metered pay-per-use (per minute), then became simplified via tiered usage plans such as AT&T Digital One Rate. “Every day low pricing” and “all-inclusive” packages can replace dynamic pricing and pay-per-use. “Loss aversion” plays a role: Sometimes you’d rather pay slightly more for a fixed price plan than take the chance of paying a lot more should usage spike on a pay-per-use plan. Of course, when the spike is revenue-generating or otherwise beneficial, Cloudonomics tells us that total cost can be minimized by judiciously leveraging pay-per-use pricing.

Additional possibilities exist, for example, program trading, risk management, trusted third-party evaluation and reporting, and rollover minutes, and causal chains are bound to happen — perishable capacity leads to dynamic pricing which begets long-term hedges.


Joe Weinman
is Strategy and Business Development VP for AT&T Business Solutions.

  1. Great post!

    Oversubscription:
    I suspect it is already in place, and Cloud providers to not have sufficient capacities in case all their customers were using their plans to the max simultaneously.

    Dynamic Pricing:
    Cloud services are not, per se, perishable goods. True, servers sitting idle are a waste, but much of the costs come from services, marketing, etc… Unlike an airplane which has a fixed route, Cloud services can be scaled much faster. The consumption is also continuously distributed amongst servers, whereas travelers are ‘attached’ to a given plane. This is not to say that Dynamic Pricing doesn’t make sense, but that perishable capacity won’t be the main driver

    Derivaties and Hedging:
    Such sophisticated financial tools can only apply to standardized goods (from AMZN stocks to Gold), so it will require Cloud services to become fully commoditized or some Cloud provider to be so prominent that a parallel market could develop for its services alone.

    Share
    1. Emmanuel..thanks!

      re: Oversubscription: a pure on-demand model can’t be oversubscribed, any more than first-come, first-served seating in a restaurant. However, yes, to the extent that reservations are being accepted, whether for the cloud or the restaurant, oversubscription is possible.

      re: Dynamic Pricing: I agree that there are subtle differences between airplanes and cloud infrastructure, although location is not irrelevant due to a variety of reasons ranging from availability to compliance regulations. However, a server-hour unused is lost forever…it can’t be stockpiled, so it certainly is perishable just like an unfilled airline seat. Service providers may be able to scale capacity up somewhat rapidly (data center construction takes a long time, but ordering new hardware for one with space can be rapid or provisioned in advance). This doesn’t solve anything though, because that capacity can’t be deprovisioned as easily. Any time there is infrastructure, there is a financial rationale for maximizing its utilization, especially given the multipliers (SG&A) associated with it, although it becomes less relevant (but not inconsequential) once value-added services are built on that infrastructure.

      re: Derivatives: agree, standardization is a key step, just like No. 2 corn, No. 11 sugar, or Hard Red Spring Wheat. And standardization in turn drives processes and organizations for measuring compliance–check out the Federal Grain Inspection Service’s standards for corn.

      Anyway, thanks for the thoughtful comments.

      Share
  2. Actually, the beginning of the new decade is still a year away.

    Share
    1. Yes. Sorry, that’s my computer science background: indexing arrays beginning at “0″ and counting time from Y2K.

      Share
  3. i would agree that technlogys are developing but you can find good use of grid computing as well..
    Control Room Temperature with Computer Script

    been trying participating and i’m pretty satisfied

    Share
  4. Joe,
    As always, great post. Not enough people have publicly applied advanced pricing theories to cloud computing; your piece pretty much sums up almost every option that was on my radar.

    You do mention the cloud provider near a stock exchange model, but a broader extension of that is the network effect, which means that cloud providers who have great “anchor tenants” will create demand for partners of the anchor tenants to reside on the same cloud. Higher demand leads to higher prices, so at the end of the day the cloud provider with the fastest (or cheapest in the intra-data-center bandwidth pricing model) access to Facebook or SalesForce and other key tenants will have an advantage.

    At the end of the day, after we’ve applied every economic and financial tool at our disposal to price the cloud, we’ll end up with the most basic situation – traffic will be routed to the cloud that can service it for the lowest cost while meeting service level standards.

    I predict that within 2 years, at least one major cloud vendor will present current pricing – which will vary – via an XML API or perhaps RSS.

    Keep up the good writing Joe!

    Share
    1. Dave,

      Thanks! I agree: Albert-Laszlo Barabasi calls it “preferential attachment” and suggests (based on extensive studies of the Internet architecture) that it leads to a power law distribution in network (or by extension, cloud) size.

      Joe

      Share
  5. Very interesting set of analogies, and in general I agree that there is no reason why many of these aspects of other on-demand services should not apply to cloud computing. There are some however that I hope the cloud computing world can stave off for as long as possible!

    Share
  6. Joe, your choice of analogy examples makes me wonder, would AT&T participate in a site that used a Priceline.com or Hotwire.com business model to sell cloud services? Where the buyer selects critiria for quality and price, and only once the purchase transaction is complete would they discover the name of the supplier.

    Given your background, are you suggesting that managed cloud services are inherently a commodity offering — similar to a domestic long-distance phone call minute?

    Share
  7. [...] Hedging Your Options for the Cloud:  In this GigaOM post, Joe Weinman of AT&T discusses a variety of economic and business models for cloud computing. [...]

    Share
  8. [...] on cloud computing are respected by many in the Clouderati, wrote a thought provoking post on GigaOM about various ways in which cloud computing pricing models will evolve. One of the them was the [...]

    Share
  9. Hi Joe

    Very interesting ideas – as usual!

    Two comments:

    1. The extent to which these pricing schemes will actually be implemented probably depends on how the value chain evolves. If vertical integration (i.e. facilities-based competition) carries the day, it’s likely to happen. If a model where wholesale IaaS providers supply retail cloud computing/SaaS providers, then these pricing mechanism are less likely to be implemented.

    Vertically integrated suppliers will obviously want to optimize utilization and implement things like dynamic pricing, oversubscription, etc. If most CC competitors are vertically integrated these pricing models will probably become inevitable. If instead many of the large carriers opt for a resell model, then the (wholesale) IaaS provider may not be able to implement these kinds of pricing schemes – as his customers/retail channels will want to offer their customer simple on-demand pay per use with guaranteed availability. In that case the wholesaler might have to absorb the risk.

    Vertical integration yields cost advantages (partly because it enables pricing and other mechanisms for optimal capacity utilization). But whether this is how things will evolve depends on whether carriers and outsourcers like IBM and HP view each other as competitors or as partners in this space. In the latter case, the pricing models you describe will come to pass only if the wholesale guys have the upper hand in the relationship – which seems unlikely to me, given the obvious advantages carriers have to become the dominant retailers of these services.

    1. Another consideration is competition with substitutes. Public cloud computing will eventually compete with so-called “private cloud” solutions. Rather than aggregate cost, the main value proposition of public cloud is reducing financial risk by minimizing fixed IT costs. The best way to do that is simple on-demand, utility style pricing. More complex pricing schemes such as the ones the airlines have implemented are generally less popular with customers. Unlike cloud computing providers, airlines can get away with this, because there is really no substitute for the type of third party service they provide – unlike in-house IT, keeping a fleet of private planes for business travel has never been a cost efficient solution even for the largest corporation.
    Share

Comments have been disabled for this post