Google Preemptible VM, AWS Spot Instances and SLA Based (Cloud) Computing

Jun 14 2015 | by Vittaly Tavor

65 Flares Twitter 0 Facebook 0 LinkedIn 65 Email -- 65 Flares ×

The AWS cloud introduced Spot instances about six years ago. That was very early in terms of public cloud adoption. This week, it was Google’s turn to take a step with their new Preemptible VMs. In addition to reducing the cost of various types of VMs by up to 30%, the new Google Preemptible VMs can save users 70% on the baseline cost of Google Cloud Platform’s (GCP) single VM. In this article will discuss the balance between the SLAs and the price of a single cloud resource within the two cloud leaders’ specific “spot” offerings as well as the motivation behind giving such significant discounts.

Cloud Vendor Motivation

Public cloud vendors took the challenges associated with traditional data center capacity planning away from their customers and put them on themselves in an even more extreme manner. One of the major risks they took by doing this was being left with unutilized hardware capacity that essentially “bites” their margins. The significant investment in an enormous amount of hardware resources encourages IaaS vendors to find ways to generate a consistently high level of data center resource utilization. Spot and reserved compute capacity are key offerings for that matter.

Amazon has a market for every offering, from books to Spot Instances. In order to get the best price, AWS users have to track spot market fluctuations and bid accordingly. Conversely, when GCP announced their Preemptible VMs this week, they highlighted:  

“Importantly, unlike other clouds’ Spot Instances, the price of Google Preemptible VMs is fixed – making their costs predictable.”

At the end of the day, the Google Preemptible VM is the same as an EC2 Spot Instance aside from the fact that it has a fixed price. While this move is not surprising, it is definitely an important milestone in the realm of the cloud computing market. It is a step towards the evolution of real competition in a market whose current leader (Amazon) is years ahead of the game. 

SLA-Based Computing

As with Spot Instances, Google Preemptible Instances come with no guarantee and can be taken away from users at any point in time. In addition to their limited 24 hour usage ability, they can be shut down with no prior notification, as opposed to AWS’ EC2 Spot Instance Termination Notice feature that was announced by Amazon at the start of this year.

The notion of having an SLA as another dimension for a single VM cost was dealt with mostly in the academic world. However an exception was made with ClusterK, an interesting startup that helps AWS users leverage Spot Instances to serve production by wrapping single instances with a “failover mechanism”. That way, if a spot instance is no longer available, the instance content automatically migrates to an On-Demand or Reserved Instance, ultimately creating an optimal price model for running application workloads on AWS cloud. What might be more interesting is the fact that Amazon recently acquired ClusterK.

Despite the appealing and seductive prices, Spot Instances are still not the top choice among users. As estimated in the past, spot use is around 5% of the whole pool of EC2  instance types. Clearly, users still look for certainty even in non-mission critical environments. Finally, SLA-based computing is great in theory, but in reality it puts a price tag on instances according to their availability, which can create an extreme level of management complexity for both cloud vendors and their clients.

Short Summary

Google’s blog announcement:

“Our combination of sustained use discounting, no prepaid lock-in and per-minute billing offers users a structural price advantage which becomes apparent when we consider real-world applications.”

In short, this is why cloud cost optimization systems exist. With these new options, cloud users need the ability to make informed decisions that serve their specific IT needs at optimal costs. They need to know when and where to host their application workloads so that performance, as well as their monthly cloud bill, won’t be negatively affected. The new offering by Google may also enable portability options that can lead to a cloud exchange market, but that can be left for a different blog post.

Take Control Over Your Multi-Cloud Deployment

Take a 20-day free trial, view your entire cloud deployment in one place and identify opportunities for greater cloud-efficiency. Obtain enterprise-wide cost transparency with customized reporting and analytics for your AWS, Google and Microsoft Azure clouds.
start free trial


Forgot Password?

No account yet? Register.

65 Flares Twitter 0 Facebook 0 LinkedIn 65 Email -- 65 Flares ×