One of the best aspects of AWS-style cloud computing is the ability to quickly scale up in times of high traffic. Reddit, for example, temporarily added 40% extra capacity to cope with last year’s Obama AMA using Amazon’s cloud service.
However, while Bezo’s behemoth may be thebiggest cloud provider, it’s far from the only option available. Microsoft arestarting to take Azure seriously, and this week Google trumpeted the advantages of Google Compute Engine with an impressive demo of their load balancing service.
In the days before Nginx, the target was 10,000 simultaneous requests: now, Google haveshown off their virtualised infrastructure scaling up to 1,000,000 per second in just five seconds.
In May, Google Compute Engine hosted the backend for the Eurovision Song Contest companion app. The TV show is watched live by 125 million people across Europe, and its engineers managed to scale the app to a whopping 50,000 requests per second.
In comparison, wrote Voellm, the load generated in this experiment was 20 times that amount – equivalent to truly enormous massive spike in traffic. The million requests per second were served by 200 single-core VMs, each receiving approximately 5,000 requests per second.
The key boast for Google is the speed at which these VMs were spun up: apparently within five seconds and without any ‘pre-warming’ whatsoever. AWS’ Elastic Load Balancing service is nowhere near as flexible – requiring both a $49-monthly premium support agreement and notification of high traffic in advance.
Equally impressive is the fact that this setup cost just $10 USD. Granted, this might go up a notch when sending real responses (the service is billed by GB of data processed in addition to an hourly service charge), but still appears to be jaw-dropping value for money.
As AWS continues to dominate the market, it’s good to see real competition and innovation from rivals – even if it results in Google controlling even more of the internet.