Securitising the cloud
As the conversation at cloud events starts to move on from ’101 – what is a cloud?’ the economics of cloud computing is beginning to become a popular theme.
Like so many things in this area there is nothing really new, just recycling of old ideas. My former boss Steve Yatko first presented his thoughts about a ‘Virtualized Resource Market’ at the Grid Today conference in 2004 (and with a bit of spit and polish I presented similar ideas at OGF20 some years later). I’m sure if we look back further there were probably some guys from IBM thinking about marketplaces for time on mainframes back in the 50s or 60s.
Typically as soon as this conversation gets started somebody tries to point out what they think is an obvious flaw – ‘how can you deal with spiky demand’ (or it’s close cousin ‘what happens when everybody gets busy at the same time’)? The answer to this falls into broadly two categories:
- At sufficient scale things get more predictable. Whilst it may be very hard to predict the usage patterns and resulting load for individual web sites/services etc. on an hour by hour or day by day basis this task gets a lot easier when dealing with large numbers. When thousands of workloads are aggregated together then you have a lot of peaks and troughs to cancel each other out. This is what securitisation is all about. If I’m an unsigned band with no track record then I have to wait with trepidation to see what’s in the tip jar at the end of a gig, whilst if I’m David Bowie I can sell my future cash flows to a bank in return for a bunch of bonds with known coupons.
- The market price regulates demand. One of the central points of a virtual resource market is to provide capacity management. Capacity becomes cheap when it is underutilised (as we see with AWS spot pricing), you may also buy capacity more cheaply if you give good notice of your need (as we see with AWS reserved instances). What we’ve not seen yet is the market rate rising when things get busy (the AWS on demand rate remains constant [in fact it drops every once in a while to reflect Moores law]), but it’s fairly safe to predict that future markets will exist where on demand pricing rises (steeply) when demand is high.
Of course it’s still very early days for the development of the cloud economy. As Simon Wardley continues to point out in his excellent presentations ‘cloud’ is simply a label that we’re applying to the trend of IT being commoditised. What we are starting to see is the emergence of a commodity market (and a fractured one at that). The virtual resource market that my colleagues and I envisaged was one where a service level agreement (SLA) was essentially a derivative contract for the underlying commodities (compute, storage and IO) – so we’ve got a long way to go yet.
