In a new twist, Amazon has again repackaged and repriced its services, this time allowing Amazon EC2 instances to be treated as if they were scheduled jobs -- at up to 10 percent off Amazon's usual prices.
SRI (Scheduled Reserved Instances) allows customers to "reserve capacity on a recurring basis with a daily, weekly, or monthly schedule over the course of a one-year term," according to the blog post announcing the new feature.
Purchasing an SRI is similar to setting up a conventional reserved instance. After the user selects a time frame for running instances, Amazon provides a list of instance types available for that segment.
To that end, using SRIs slightly resembles using EC2 Spot instances. Instead of bidding for capacity, as one would with a Spot Instance, customer bids for one of a limited number of usage slots available in a given window.
SRI usage revolves around batch-operation scenarios -- that is, a company performing tabulations for a given month on a specific weekend or a compute job for a 3D rendering that needs to be performed overnight. Such jobs only require an instance to spin up, run, deliver their results into an AWS storage instance, then shut down.
Why use SRIs instead of simply hand-scheduling those jobs? Convenience and cost. SRIs are 5 percent less expensive than standard on-demand instances during peak hours, and 10 percent cheaper during off-peak (weekends).
One major drawback to SRIs: You have to make upfront commitments to usage. You can book daily, weekly, or monthly schedules, but you have to book them over the course of a one-year term -- and you have to book at least 1,200 hours of utilization for the term, or an average of 3.28 hours per day across the course of a year. Unused capacity during the course of the term does not appear to be refundable, and you can't book the start of an SRI term more than three months before the start of the term.
The premise of AWS, and indeed the cloud itself, is on-demand, part-time computing. Amazon has pushed that envelope and found ways to further decouple the workloads from the instances. For example, AWS Lambda turns snippets of Java, Python, and Node.js code into on-demand services that don't require a whole AWS instance to run.
Amazon eventually plans to cross-integrate SRI with Lambda, although it's not yet clear the two will complement each other. In one likely scenario, Lambda instances would only run during predefined times and at a discount rate -- cost-effectively making APIs that are only needed occasionally. Amazon also plans SRI integration with its Auto Scaling and CloudFormation services, so SRI instances could theoretically be scaled to meet changing demand across time or controlled programmatically via templates.