Google and Amazon are duking it out for the cloud services pot, trading innovative blows back and forth in a fight that’s all good news for developers.

In a Google Cloud Platform blog post this week, Google performance engineering manager Anthony F. Voellm laid out a sustained way to scale a cloud application deployment to handle a million requests per second, within five seconds after setup.

Oh, and the setup and testing cost only $10.

Google Cloud Platform services made this possible using Compute Engine Load Balancing, which avoids convoluted DNS load balancing by offering simple APIs and allowing a single IP address to serve all traffic. Compute Engine Load Balancing also eliminates 404 errors by detecting unhealthy requests, and by adding and removing requests.

Compute Engine Load Balancing

The figure shows Google’s Compute Engine Load Balancing setup for cloud application deployment.

Voellm starts with an empty Compute Engine project ending with 456 provisioned cores for the load generator and Web servers, and one load-balanced IP address actively processing 1.016 million requests per second. To generate the load, they ran the open-source tool curl-loader, with 16 threads and 1,000 connections. Each instance of curl-loader ran the same configuration to generate roughly the same number of load-balancing requests.

The scaling itself involved running Apache v2.2.22 on Debian 7.1, evenly distributing each of the 200 back-end Web servers with approximately 5,000 requests per second.

Amazon Web Services’ comparable service, Elastic Load Balancing, offers similar scaling… if you sign up for a premium support agreement. Google’s new method promises instant scaling without funneling requests through a support team, for less money than buying a movie ticket.

Your move, AWS.

Read more about Google’s Compute Engine Load Balancing here.