For cloud computing aficionados, the use of spot instances is a potentially important strategy. And it’s probably time for cloud newbies to get educated on this phenomenon.
Launched six years ago by Amazon Web Services (AMZN), spot instances let customers bid on unused computing capacity and benefit from potentially huge discounts—possibly up to 90% compared to its pricier on-demand instances. (Experts say discounts more typically range around 40%, but hey, that’s not nothing.)
Amazon pretty much had this market to itself until last week when Google launched what it called “preemptible VMs” as in virtual machines, the term for software that imitates dedicated hardware. These resources can be had for up to 70% discount off of Google’s (GOOG) standard compute pricing. Overall, Google’s service is pretty similar to Amazon’s spot instances except that Google makes its instances available at a fixed, discounted price so the customer doesn’t need to keep checking the spot market for fluctuations, said MSV Janakiram, a cloud computing consultant who probably knows more about this market, which he wrote about here, than any other human.
The same day Google posted its news, Amazon announced a new tool that lets customers launch and manage thousands of spot instances with one request without having to write any code.
Janakiram said he has several large clients who use spot instances now to cut their costs, especially for batch computing jobs that do not require immediate processing. In these cases, spot instances or preemptive VMs can be a God send, in his view. And use of these resources is only going to grow as companies get smarter about saving money in the cloud.
A ticketing agency that generates a ton of computer logs over the course of business uses AWS spot instances very effectively, he said. “At the end of the week they want to know where their users are coming from so they take those workloads run them on spot instances,” he said. Spot instance price typically falls on weekends so running those jobs in off-hours is very economical.
In addition, big companies with sophisticated IT chops can schedule jobs to follow the cheapest spot instances around the clock and around the world. “If you’re a company in Asia and your peak traffic comes in a window that is not a peak time in Europe, you may run it there for a lot less money,” he said.
A word of caution: The application has to be designed to be able to take advantage of this capability. Legacy enterprise applications will not be able to follow the sun in this way.
Descartes Labs uses both Google and Amazon (AMZN) options for different applications, said Co-founder Tim Kelton. Amazon spot instances are used primarily for graphics-intensive jobs because Amazon offers a special type of instance crafted for those workloads, while Google does not. (At least yet.)
“We use Google preemptible instances for scenarios where we have more flexibility on processing a workload but prefer to do it at a much lower cost or burst workloads above normal capacity estimates,” Kelton said via email.
The biggest difference he sees is that Google offers a fixed discount rate while Amazon requires a bid price from the customer for a particular instance type in a particular data center.
And, he noted, with Amazon “you can lose your spot instance, not because AWS no longer has excess spot capacity but rather because another AWS user bid a higher price for the instance you’re using. ” Google’s instances last for 24 hours, which he finds useful for processing large distributed workloads while AWS spot instances can last as long as excess capacity remains available and no one else bids up the price.
So these delivery models right now are good for the sort of jobs that can wait for a window of opportunity. But that too will likely change.
Amazon’s acquisition of ClusterK last month could make it easier for companies to put even their most important jobs on cheap spot instances. “ClusterK has an algorithm that will shuffle the job between spot and on-demand instances,” Janakiram noted. It constantly checks for availability of instances among the various AWS zones and customers don’t have to worry about manually putting their jobs on the cheapest machines, he said.
“If the price drops in one zone, ClusterK will bring up a machine in that zone and re-factor and rebalance the whole load.”
All of this, of course, is good news for customers—at least customers who are cloud savvy. As Amazon, Google and Microsoft (MSFT) keep trading price cuts and higher level services in competition with each other (and a dozen or more other providers) business customers can reap the rewards.
And now that Google and Amazon have weighed in with new discounted compute, can Microsoft be far behind? Tick tock.