Cloud Innovation – Temporary Jobs

Call them Spot Instances, Preemptible VMs or whatever, but short-lived, non-mission-critical, temporary cloud compute jobs are emerging as a second front in the cloud compute wars.
Big cloud players can only reduce prices so much. Brand-new, major innovations, however, are rare and far between. Providers such as Amazon Web Services Inc. (AWS), and Google Cloud Platform are looking for new ways to compete. For example, they offer unused capacity for ephemeral computing jobs.
Google announced this week the Google Compute Engine Preemptible virtual machines, a beta program that offers regular compute instances at a lower cost. However, they can be shut down at any moment. Preemptible VMs can be shut down in 30 seconds and will only run for 24 hours.
Paul Nash, a Google executive, stated that while it may sound disruptive, it makes them a great choice to distribute fault-tolerant workloads that don’t require continuous availability of any one instance. We are able offer them at a significant discount to normal instances because we don’t guarantee indefinite uptime. Fixed pricing starts at $0.01 an hour.
Industry pundits were quick to dismiss the usual titt-for-tat responses from AWS and Microsoft Azure as the typical response. However, AWS had already begun the work for improving Spot Instances with the acquisition of ClusterK, a specialist in Spot Instance bidding.
AWS also announced a new tool that will simplify the management Spot Instances in fleets of thousands of people on the exact same day as Google’s announcement.
AWS was quiet about ClusterK’s acquisition. However, it confirmed the purchase of the company that helps customers save time by managing the bidding process to find the best Spot Instances deals. Prices can vary, unlike Google’s counterparts.
The AWS site states that Spot Instances allow users to set their own prices for Amazon EC2 computing power. You simply bid on spare Amazon EC2 instances, and they will be run whenever your bid exceeds current Spot Price. This fluctuates in real-time based upon supply and demand. The Spot Instance pricing model is a complement to the Reserved Instance and On-Demand pricing models. It often provides the most cost-effective computing power, depending on the application.
ClusterK leverages proprietary technology to find the best deals using a special algorithm. It also guarantees uptime so mission-critical jobs are not interrupted by the Spot Instances.
AWS announced this week that it has a new tool to manage multiple Spot Instance deployments.
AWS exec Jeff Barr stated that Amazon EC2 Spot Fleet API allows users to manage thousands of Spot Instances using one request.
Barr stated that EC2’s Spot Instances model is now even more useful. He added that a new API allows you to launch and manage a whole fleet of Spot Instances from one request. A fleet is a collection or Spot Instances that work together in a distributed application. A fleet could be a batch job, a Hadoop workflow or an HPC grid computing job.
Barr stated that AWS customers used to write their own code to manage deployment of these fleets. This is no longer necessary.
Instead, a single API function (RequestSpotFleet), does all the work for you. Simply specify the fleet’s target size, a bid price per an hour, and tell Spot which instance types you wish to launch. Spot will search for the lowest-priced spare EC2 capacity and work with you to reach and maintain the fleet’s target capacities. One call does it all, as they claim ….”
Barr stated that this was the first release of a brand new product.