Google's Preemptible Offers Virtual Machine-for-Hire Service on the Cheap

Thursday Jun 18th 2015 by Paul Rubens
Share:

Willing to sacrifice reliability to save some cash when it comes to server virtualization? Google's Compute Engine Preemptible service may -- or may not -- be your best bet.

With virtual machines, as in life, you get what you pay for. So if a service provider is offering very cheap virtualized servers in the public cloud, it's a pretty sure bet that they won't be reliable ones.

But sometimes cheap is just what you are looking for, and reliability be damned. Virtually Speaking And it's that kind of thinking that is behind a new cloud-based virtual machine-for-hire service from Google, called Google Compute Engine Preemptible (GCEP) Instances or virtual machines.

And cheap the service most certainly is — with Preemptible virtual machines priced for hire from as little as 1/2 cent per core per hour — and often less than half the standard rates.

But reliable? Not so much. The main "feature" of the service is that with Google's Preemptible any virtual machine can be shut down at any time with no warning. Poof – gone!

VM Ticking Time Bombs Waiting to Disappear?

In fact, Google's Preemptible virtual machines are ticking time bombs waiting to disappear in a puff of logic. That's because if they have a maximum life span of 24 hours, assuming they don't get shut down sooner, they are guaranteed to pop out of existence as the clock strikes 12. Or more accurately, 24.

While the service sounds like some sort of April Fool's joke initially, after a moment's consideration it's clear that Google may be on to something here. VMs that can disappear without warning work fine for distributed, fault-tolerant workloads that don't require continuous availability of any single instance, as Paul Nash, Google's senior product manager, points out.

In fact, servers can and do fail with little or no notice at any time, so for tasks that are designed to tolerate this, what's the problem?

And of course no one is forcing you to do all your computing on this type of virtual machine. For example, there's no reason why you can't set up a Hadoop cluster in Google's cloud and specify that, say, 25% of the servers in your cluster should be Preemptible VMs.

Recommended for You







Paul Rubens is a technology journalist and contributor to ServerWatch, EnterpriseNetworkingPlanet and EnterpriseMobileToday. He has also covered technology for international newspapers and magazines including The Economist and The Financial Times since 1991.

Follow ServerWatch on Twitter and on Facebook

What's the Worst that Can Happen?

In a scenario like that, what's the worst that can happen? If all 25% disappear, your Big Data number crunching will take a little longer than if none of them had. But the upside is that it will only cost you marginally more for the chance of the task being completed considerably faster. Which may, or may, not suit you.

In fact, Google's Chrome security team runs its Clusterfuzz tool to perform non-stop randomized security testing in the cloud against the latest code in Chrome running on thousands of virtual machines, according to Nash.

Having more compute power means they can find (and then fix) security bugs faster. Using Preemptible VMs, they were able to double their scale while decreasing their costs, claims Nash.

The thinking behind the service is that Google inevitably has spare, unused compute resources in its data centers that it may need to claim and put to work at any time to cope with increases in demand, server failures and a multitude of other possible reasons.

So rather than let these resources sit idle, the new service allows Google to put them to useful work for customers while still being available at zero notice if the company needs them for any other purpose.

Server virtualization is all about efficiency through increased utilization, and what this does is take it a step further: increasing cloud compute resource utilization while still maintaining the necessary free headroom.

Customers win, and Google wins. All in all an excellent way to go.

Recommended for You







Paul Rubens is a technology journalist and contributor to ServerWatch, EnterpriseNetworkingPlanet and EnterpriseMobileToday. He has also covered technology for international newspapers and magazines including The Economist and The Financial Times since 1991.

Follow ServerWatch on Twitter and on Facebook

Share:
Home
Mobile Site | Full Site
Copyright 2017 © QuinStreet Inc. All Rights Reserved