Thursday 24 February 2011

Cloud infrastructure pricing and billing models

Previous posts here have considered the costs of both buying and building storage in the cloud. Based on our work so far, we estimate that is possible to build an academic cloud infrastructure (storage and compute) that is both sustainable and that can be offered in broadly the same price 'ball-park' as, say, Amazon Web Services.

Of course, the proof of the pudding will be in actually building it! Furthermore, how compelling such an offer might be to the academic community is not just down to price. We also have to think about the nature of the technology on offer, the service level agreements that we are able to provide, the associated support that might be offered, and the kind of pricing and billing models that we can put in place.

This post considers (admittedly at a very high level) the last of these - pricing and billing models. The intention is not to provide any answers at this stage. Rather I'm interested in what sorts of pricing and billing models would be attractive to the consumers of academic cloud services... in an ideal world (I have to add the 'ideal world' bit because, clearly, there may be tensions between what consumers want and what providers are able to offer in a cost-effective way - there will likely be some trade-offs to be made).

To frame this question further, let's think about some of the different parameters that might be considered in this area. Firstly, on pricing... there seem to be two options here:
  • Pay as you use (essentially the Amazon model where you are charged under various different categories for the amount of resource that you use) vs. Plan (where you commit to a certain amount of resource, paying extra (probably at a premium) if you go over your allowance. GoGrid provide an example of this kind of pricing model).
Of course, there are some subtleties around these options. Amazon, for example, offer both Reserved Instances (where hourly rates are lowered by committing up-front to using the resource for either 1 or 3 years) and Spot Instances (where you bid for unused resource against a fluctuating hourly price). One might refer to these more generically as Short-term commitment vs. Long-term commitment and Fixed-pricing vs. Bid-pricing.

Reserved Instances allow the provider to plan their capacity more accurately (hence the lower prices - my understanding is that, from a consumer perspective, an Amazon Reserved Instance breaks even if it is used for more than 30% of the time). Spot Instances soak up otherwise unused capacity, again meaning that the provider can offer better value for money.

There's also the issue of whether providers offer any kind of free-tier, as Amazon currently do, meaning that consumers can get going at little or no cost but, ultimately, driving up consumption for the provider. Again, one might refer to this generically as Introductory tier pricing vs. Flat pricing.

On billing models... there seem to be two aspects to consider:
  • Credit card vs. Invoice.
  • Up-front payment vs. Pay in arrears.
To draw an analogy with mobile phone tariffs for a moment... pay-as-you-go is (typically) a combination of what I'm calling here 'Pay as you use' and 'Up-front payment' while contract is (typically) a combination of 'Plan' and 'Pay in arrears'. (Note: I really wouldn't want to wish the complexity of mobile phone tariffs on anyone!).

My understanding, from discussions at conferences and so on, is that universities (at the institutional level) are keen to pay by Invoice rather than Credit card, that they like to Pay in arrears and that they lean towards the predictability of a Plan rather than the open-ended nature of Pay as you use. Note: individual researchers may see things differently.

So, first question... is this categorisation into:
  • Pay as you use vs. Plan
  • Up-front payment vs. Pay in arrears
  • Credit card vs. Invoice
  • Short-term vs. Long-term commitment (a la Reserved Instances)
  • Fixed-pricing vs. Bid-pricing (a la Spot Instances)
  • Introductory tier pricing vs. Flat pricing
useful? What have I missed?

Secondly... what is attractive to you (either as an individual or as an institution)?

I have to say that our thinking, to date, leads us towards sticking close to the Amazon pricing model (not least because it is a widely understood and accepted way of thinking about things), ignoring discount for long term commitments and bid-pricing (at least in the short term), and allowing payment in arrears based on a saved credit card and/or invoice (probably).

Introductory tier pricing is problematic for the provider (particularly so where the infrastructure is limited) because it is difficult to manage. Further, it tends to encourage a mindset that stuff can be made available for free when the reality is that it can't. Which is not to say that introductory tier pricing can't be done - just that its consequences need to be thought thru.

Note that all of this is just theory at the moment. It is thinking that will (primarily) inform a FleSSR project deliverable on cloud business models. As such, I welcome any and all views. Thanks.

UCISA Cloud Computing seminar

Both David Wallom (OeRC) and Matt Johnson (Eduserv) spoke at the recent UCISA Cloud Computing event in Loughborough.

Their presentations are available from the event website:Some general thoughts on the event by Andy Powell are also available on eFoundations.