Imagine you're taking a train from Washington, D.C. to New York City. You have a choice of four fares. First, you...
could buy a first-class ticket for $300, which secures you an assigned, first-class seat. Second, you could buy a coach ticket for $200, which lets you find a seat in the coach class after boarding. Third, you could wait on the platform until all ticket-holding passengers board, and then bid on any remaining seats, looking for discounts. Finally, you could book an unoccupied seat at a deep discount, knowing that you could be tossed off if the passenger who originally purchased that seat comes on board.
That third option represents the idea behind Spot Instances in Amazon Web Services Elastic Compute Cloud (AWS EC2), while the fourth option represents Preemptible VMs in Google Compute Engine. Both products offer deep discounts on machine instances for organizations willing to use cloud computing based on available or spare compute capacity. While there is no guarantee you'll get the compute resources you need, an AWS Spot Instance or Google Preemptible VM could be the cheapest way to use the public cloud.
But before deployment, it's important to weigh the pros and cons of these cloud instance types.
The pros of an AWS Spot Instance, Google Preemptible VM
An AWS Spot Instance lets users bid on spare AWS EC2 machine instances, and typically provides deep discounts on compute capacity when compared to AWS On-Demand instance pricing.
Google Preemptible VMs provide a similar service; they allow users to create a machine instance from spare capacity in Google Compute Engine that is preempted or terminated after 24 hours -- or sooner, if Google needs those resources to perform other tasks. Google's Preemptible VMs also sell at a fixed price -- which is up to 70% cheaper than regular instances, according to Google -- while AWS Spot Instances go to the highest bidder.
The pros of both options come down to money. Large organizations on tight budgets, or small businesses, could find that either an AWS Spot Instance or a Google Preemptible VM could allow them to tap into the benefits of a public infrastructure as a service at a more affordable price.
Another advantage is application throughput. With AWS Spot Instances, for example, you can scale compute-intensive applications, such as big data analytics, for as much as 50% to 90% less than the cost of doing so on an On-Demand instance, according to AWS. As a result, you can get more compute capacity for the same amount of money, if you use AWS Spot Instances with the same budget as On-Demand instances.
The team of AWS EC2 instances
With a wide selection of Amazon Web Services EC2 instance types available, AWS meets a growing range of application and workload requirements for various levels of memory, storage, CPU and networking capacity.
The danger of rushing in
Despite their benefits, there are disadvantages to using an AWS Spot Instance or Google Preemptible VM.
The most obvious drawback of these services is the inconvenience of not being able to get the capacity you need, when you need it. Traditional IT teams are accustomed to gaining access to the machines they need on-demand. These approaches, based on spare capacity, could cause a bit of culture shock.
It's also possible to spend more money on these instance types, if other users bump your spot at a critical juncture in a project, and your team is forced to wait. Consider the cost of delays if you can't use, or can't continue to use, the instances you need to run critical or even semicritical applications. For example, AWS Spot Instances may save you $10,000 a week, but waiting six hours for an instance may cost you $50,000 a week in personnel hours.
So, are these options right for your organization? If your requirements are flexible and potential project delays are acceptable, an AWS Spot Instance or Google Preemptible VM could be a way to use cloud compute instances at huge discounts. However, many people want the assurance that they have their own, dedicated place on the train; you have meetings to attend in the city, after all.
How to cut costs using Spot Instances
With Spot Instances, find the sweet spot of server configurations
Using Preemptible VMs, Spot Instances for HPC