AWS continues to make its Reserved Instances more flexible, as the company faces increased pressure from Google,...
which says it offers a more streamlined approach to discount services.
Amazon Elastic Compute Cloud Reserved Instances provide significant savings over On-Demand purchases, but a commitment to a set capacity ahead of time can spawn a host of complications. AWS has addressed some of those concerns over the years, and it took another step this week by adding the ability to segment those prepurchased compute resources to better utilize resources in real time.
AWS' Reserved Instances options have greatly expanded since the service was launched in 2009. Additions include scheduled availability times, moving instances across an availability zone and a marketplace to buy and sell unwanted instances.
Last September, AWS added tools to spread Reserved Instances across multiple availability zones within a region or to resize a VM to a different instance family. Families are Amazon's nomenclature for groups of instance types that are general-purpose or tailored for CPU, memory or storage. Customers can now break those regional AWS Reserved Instances and Convertible Reserved Instances into multiple, smaller VMs within the same family.
In a blog post intended to explain the update, AWS chief evangelist Jeff Barr cited an example of a customer with a c4.8xlarge Reserved Instance. Under the new policy, that instance could be broken into one c4.4xlarge and eight c4.large instances, or any other combination that fits within that original capacity purchased in that instance family. For spikey workloads, customers also can scale up a smaller VM, if needed, and only pay for the on-demand pricing when the larger instance size is in use.
The new capability is limited to Linux and Unix Reserved Instances, and it's available now.
Customers with smaller workloads typically can manage Reserved Instances without problems, but it can be a real challenge for users with more dynamic architectures with multiple regions, operating systems and instance types -- so much so that they may require a dedicated IT staffer to manage them.
"If you want to get the most efficient and best tech, you have to be watching your instance usage and RI [Reserved Instances] utilization every hour, which obviously isn't too realistic. So, you have to have a strategy and make sure you set yourself up for success," said John Jun, vice president of service delivery at Logicworks, an AWS managed service provider in New York.
This new feature will help manage Reserved Instances at scale, so heavy users don't have to constantly resize, modify and patch, Jun said. He oversees the Logicworks' Reserved Instances internally and for customers, and the string of updates to the program over the past six months has made a big difference.
"[Managing Reserved Instances] was a real pain in the ass a year ago, but it's gotten much easier," Jun said.
Google pushes price competition -- again
One of Google's biggest contributions to the infrastructure-as-a-service market in its initial forays was to pressure AWS and Microsoft Azure to drop prices. The hype around per-instance pricing has dissipated over time, but Google has found a new way to put the focus back on costs.
Google Cloud Platform already offers per-minute billing and custom machine types. There's also automatic Sustained Use Discounts of up to 30% for instances that run continually, with more savings available the longer the VMs are in use.
Last week, Google added Committed Use Discounts, which provide savings of up to 57% if a customer agrees to buy a fixed monthly capacity for one to three years. That program is targeted at enterprises that want to plan their expenses on an annual basis.
John Junvice president of service delivery, Logicworks
Cloud pricing, in general, is fairly complex because of the nature of pay-as-you-go pricing, but AWS' Reserved Instances, in particular, have become more complex as the range of options has increased, said Owen Rogers, research director at 451 Research. The attraction to the new Google program is the ability to buy a quantity of CPU and memory in advance and consume it as you see fit.
"The new model enables liquidity of cloud resources far more easily than the AWS model," Rogers said.
Capacity planning is one of the biggest problems in the cloud, as customers struggle to build new applications, said Rob Harrop, CEO at Skipjaq, a London-based startup that builds software-as-a-service-based performance optimization on AWS. Customers often don't understand how to accurately size their VMs for performance or for end-user reception, so they often overprovision the instances.
Amazon will work with customers to avoid overprovisioning once they spend around $5,000 a month, Harrop said. But with AWS Reserved Instances, if a purchased instance ends up being the wrong size, a customer would have to go to the secondary market or just stick with what was bought.
"At least, now, you can move up and down," Harrop said. "It's just obvious ... Google inherently has more ways to pick and choose your configuration."
Part of the difference between the two discount models could be how they architect their clouds. AWS is believed to have a more VM-centric approach to its data centers. Google uses containers and treats its data centers as one large supercomputer, so customers can slice up servers as needed.
Whether or not the AWS underlying infrastructure configuration and pricing are connected, some customers see advantages in the Reserved Instance model. For example, Google says its service is designed to avoid upfront costs and provide maximum flexibility, but Jun said there's a real benefit in the ability to prioritize in a region or availability zone through a particular Reserved Instance type and time on AWS.
And Google's pricing model differentiation could be short-term -- any other provider could simply copy the process, Rogers said. AWS also could maintain the benefits of the Reserved Instance model, but add automatic VM optimization to remove some of the management complexity.
AWS declined to comment for this story.
Trevor Jones is a news writer with SearchCloudComputing and SearchAWS. Contact him at firstname.lastname@example.org.
Make an economical choice with Reserved Instances
Take a hard look at the big three public cloud providers
Free tiers of service vary cloud to cloud