Google boasts of giant Compute Engine demo, boosts uptime with live migration.
Google Compute Engine, the company’s infrastructure-as-a-service cloud that competes against Amazon Web Services, is trying to take reliability and scale to the extreme.
Last week, the company said it was able to serve “one million load balanced requests per second” with a single IP address receiving the traffic and distributing it across 200 Web servers. Each of the million requests was just “one byte in size not including the http headers,” Google Performance Engineering Manager Anthony F. Voellm wrote in a blog. It’s thus not representative of real-world traffic, but the simulation shows that Compute Engine should be able to let websites absorb big bursts in traffic without shutting down.
According to Google, the test showed the load balancer was able to serve the aforementioned one million requests “within five seconds after the setup and without any pre-warming.” The test ran for more than seven minutes. “The 1M number is measuring a complete request and successful response,” Voellm wrote.
The Google Compute Engine load balancer is an option available to customers of the cloud service. The simulation would have cost $10 based on the service pricing, Google said.
To test the service, Google said it used curl-loader, an open source tool that simulates application load and behavior. Curl-loader was run from 64 virtual machines, which directed the load to “a single IP address, which then fanned out to the Web servers.”
The 200 Web servers each received about 5,000 requests per second, adding up to a million.
“Compute Engine Load Balancing distributed the load by using a tuple of source address+port, destination address+port, and protocol,” Google said. “Each Web response was one byte in size, not including the http headers. All of this was configured from an empty Compute Engine project. To reproduce the data yourself, you can use the following Gist. This entire setup and test cost just $10 USD!”
In addition to the big load balancing simulation, Google declared yesterday that Compute Engine is now “generally available.” Compute Engine was already opened to anyone who wanted to start using it in May of this year, so the change is mainly a naming exercise akin to finally lifting Gmail’s beta tag so many years ago. Compute Engine went from newborn to full-fledged service in just a year and a half.
More importantly than “general availability,” Google announced new features that will improve reliability. These include live migration of virtual machines so that they keep running when Google performs data center maintenance and automatic restart of VMs in the event of a data center failure.
This is an important improvement, according to Gartner analyst Chris Gaun (who has been testing the Google cloud). Google used to terminate virtual machines during scheduled maintenance times, meaning that customers had to move the virtual machines or lose them, he told Ars. Now, customers don’t have to worry about that annoying possibility.
Additionally, Google lowered the prices of standard instances by 10 percent and lowered the price of persistent disk storage by 60 percent per gigabyte. Google is also now supporting “any out-of-the-box Linux distribution,” instead of just Debian and CentOS. Additionally, Google unveiled 16-core instances with up to 104GB of RAM.
Google still lags Amazon in enterprise features and the number of data center regions and availability zones, but it is making steady improvement, Gaun said. While Google provides service from data centers in the US and Europe, Amazon Web Services is in the US, Europe, South America, and Asia Pacific region.