Amazon have announced that EC2 users can modify the instance type of reserved instances (where users pay a lower rate for compute in exchange for a longer-term commitment).
This is achieved by the use of interchangeable units within each instance family (m1, m2, m3 or c1), with a small instance counting as one unit and binary size increments up to 8xlarge (64 units). The move will allow AWS users to have greater flexibility in how they consume and pay for capacity.
Flexible sizing comes in addition to the ability to move EC2 reserved instances that was announced in September, which allows portability between availability zones within a region and movement from EC2 classic to Virtual Private Cloud (VPC). Since March 2013, there is no longer any premium for using VPC, though associated services like IPsec gateways and NAT instances do incur additional charges.
It’s worth noting that only the ‘heavy utilization’ reserved instances come with an obligation to pay for usage, whether an instance is running or not. The light and medium utilization offerings are for customers who run instances a lot, but not all the time. In this case ‘utilization’ has no connection to how hard an instance is working once it’s running. The only instance type that’s offered for occasionally busy workloads is t1.micro, and that’s not on the reserved instance menu.
Whilst the ability to modify sizing allows greater flexibility in instance deployment, it’s still fundamentally binary. So competitors like Verizon’s new public cloud may still be able to attract customers by offering bespoke sizing for instance types. Amazon have also stuck with per-hour billing across the board, and haven’t responded to Google and Microsoft’s moves to per-minute billing. Finer billing resolution is particularly useful to customers running short-duration workloads, but also offers a small overall discount for those with longer duration usage.
The increased flexibility in sizing and location should prove useful to cloud brokers like Cloud Options (formerly Strategic Blue), as it will make it easier to resell allocations from one customer to another. The brokerage market is starting to heat up, with Deutsche Börse launching it’s Cloud Exchange (DBCE) and the Chicago Mercantile Exchange (CME) partnering with 6fusion on their Infrastructure as a Service (IaaS) Marketplace.
A more liquid market for IaaS will help customers find the best match of pricing and commitment, but there is no sign of a cloud price war. Service providers like Microsoft and Google have pinned their instance pricing to Amazon’s, and all of the providers suffer from poor elasticity of supply. It just takes too much time and capital to build out fresh capacity. At a recent interview with Om Malik, Joyent founder Jason Hoffman commented on instance prices being ‘discounted 83% relative to their 2009 number’, whilst ‘internally the cost has been discounted 98%’. IaaS has become a higher margin business over time as lower prices for users have failed to track Moore’s law driving reductions in supplier costs.
Amazon has also launched its Second Annual EC2 Spotathon, a competition to find interesting use cases for spot instances. Spot pricing auctions are used by cloud service providers like Amazon to sell any remaining capacity that’s available after what’s used by reserved instances and on demand instances. Users with workloads that aren’t time sensitive can take advantage of lower spot pricing in order to consume infrastructure at a much reduced cost. Last year’s winners were PiCloud, a PaaS for scientific high performance computing, and Princeton Consultants’ OptiSpotter application for back testing high frequency trading algorithms.