David Chappell


Get the Feed! Subscribe

Amazon Reserved Instances and Cloud Capacity Planning  
# Saturday, March 21, 2009
Amazon's announcement of discounted pricing for EC2 customers who make long-term commitments is welcome news. Cheaper is better.

One way to look at this change is to see it solely as a price reduction, something focused on addressing customer concerns. But adding reservations also addresses Amazon's own concerns.

Here's the issue: Customers want to see EC2 and similar public cloud platforms as an infinite pool of compute resources. Rather than building out your own data center to handle peak loads, you can instead count on your cloud provider always to have enough capacity. The providers themselves encourage this view, since it's one of their strongest economic arguments.

But can a cloud platform really provide infinite compute resources? Of course not. Relying on a cloud provider for peak computing resources doesn't make the capacity planning problem go away: It just moves it out of your data center and into the cloud. And the people who run cloud data centers now face this problem in a big way. As more firms depend on them, they need to predict more accurately what their aggregate demand will be.

With the original EC2 approach of VMs on demand, this is essentially impossible. While a cloud capacity planner might make some inferences based on previous demand, number of new customers, and other variables, accurately predicting the future is tough when anybody can request a bunch of new VMs with no warning.

Reservations make it easier. By knowing more up front about a customer's likely usage, cloud platform planners can make better decisions. Since having too much capacity hurts the cloud provider's profits, while not having enough capacity at times of heavy demand is really bad for business, knowledge about future usage has enormous value.

So yes, Amazon is lowering prices. At the same time, though, they're improving their own ability to make intelligent investments in their business.

As the innovator, Amazon is once again defining our expectations about what a public cloud platform is and how it should be priced. Because reservations are so valuable to the people who run a cloud platform, expect the idea to spread.

1 comments :: Post a Comment



Yes agreed but I did an analysis of a recent posting in January09 by Chris Fleck's Blog on AWS reserve instance. It appears the economic value are only applicable to certain low to medium usage workloads which from the scenario case study appear to be around the intermittent loads of 8 hours or less per day. The 24x7x365 3 year example is still expensive relative to purchase cost for predictable loads. The hybrid burst scenario is much more compelling economically giving results that suggest a lower cost than a on-premise purchase. This is predicated with a switch capability to be able to put variable loads into to the cloud. I believe the Microsoft Azure model does this and is looking equally very compelling potentially.

The threshold of buist (both over capacity or selected on-demand capacity) I have coined "the economic click barrier" to cloud computer CAP and OPex adoption. There is also another clock barrier the absolute cost of OPex versus purchase cost on-premise which is where the cost of a short temr Dev/test environment is far cheaper in a cloud scenario by the order of 1:50 or more such as a 4 week project using a high end CPU server configuration capacity for just 40 hours x 4 weeks approx to $600 only.

BTW really enjoyed the Feb 09 white paper publication on Microsoft Azure, excellent document.

Best Regards
Mark Skilton
Cloud Computing Evangelist, CSC

Post a Comment

<< Home