Cloud 101 - Lesson 3.6 - The Elephant In The Livingroom: Unused Capacity
Lesson 3, "Behind The Scenes in Determining the Costs of Cloud Computing" continues with this discussion of how unused capacity affects the cost of cloud computing. This lesson builds on the previous ones. Please see the overview in Lesson 3.1 if you have not already read it.
So far, we’ve discussed how the cost to the cloud computing vendor is affected by how their application instances are allocated on the vendor’s servers. However, there is an even larger factor that affects the vendor’s costs, which is how many unused servers they keep in their inventory. For example Amazon EC2 rose to fame with their now-famous example of how they digitized nearly 100 years of microfiches using 100 virtual servers in one day. Now, customers routinely use EC2 to provide thousands of machines for short periods to perform medical or scientific calculations. However to provide these burst of computing, Amazon has to keep a large inventory of unused servers around, by some estimates as much as 66% of their total on the average. This of course adds greatly to their costs for data center space, equipment cost, and repairs, even if the servers are turned off when not in use. This isn’t generally discussed by cloud providers, but if you examine the quarterly reports of the cloud providers that are public companies, you may be able to see some interesting facts. One would expect that over time, these costs would have to be passed on to the customers, even if the vendor keeps the prices low for a while to buy market share. In other words, you can expect cloud computing costs to rise over time as vendors establish themselves.
While the bulk of business and web computing varies in load over time, the variation is typically constrained and in the range of two to three times the minimum usage, allowing for an average load across many customers that varies far less than what Amazon must build their datacenters to accommodate. If a cloud vendor is willing to give up on taking business from customers who want thousands of servers for short periods of time, they should be able to manage their costs by keeping a much smaller inventory of unused equipment in stock, and in turn passing the savings on to their customers.
For vendors that choose to supporty widely varying loads, the latest "innovation" is allowing customers to reserve capacity so that they can be sure the computer power will be there when they need it (which Amazon doesn't guarantee, otherwise.) Especially in the case of prepaid reservations, this allows the cloud vendor to get help from the customers in purchasing hardware which otherwise would simply be kept around at a loss.
TrackBack URI for this entry
