As the person responsible for running the world’s largest public cloud, Amazon.com Inc.’s James Hamilton knows a thing or two about delivering technology services at scale. The industry-renowned engineer returned to theCUBE at the retail giant’s recent re:Invent summit earlier this month to provide a rare glimpse into the inner workings of the global infrastructure juggernaut he helped build.

Amazon is hardly the first to have implemented the model of selling hardware resources on-demand over the web, but the fact that it has succeeded where others failed is a feat Hamilton credited to a unique approach to networking. Early infrastructure as a service providers of the 1990s struggled to deliver their services to the enterprise securely and efficiently but were frustrated by expensive and inefficient networks. The retail giant has built its cloudy empire on making infrastructure rapidly accessible to organizations at cutthroat rates thanks to scale as network efficiences.

That flexibility didn’t come easily. To meet the requirements of its more than one million users, the globally distributed Amazon Web Services (AWS) environment has “private links between all the regions for better costs, better availability and lower latency,” Hamilton told theCUBE hosts John Furrier and Stu Miniman. “And at the data center level, we run all custom Amazon-designed gear and protocol stacks that allow us to continuously drop costs.”

Amazon has managed to buck the trend of the broader networking world, which has not seen the same kind of price declines as other technologies. Hamilton sees room for change. “Networking needs to be the same place where servers went 20 years ago and that is a Moore’s Law curve,” he said.

Operational efficiency is a central priority for Amazon, but it’s only one of the several factors behind the success of AWS. Another important element is that the company typically maintains multiple data centers in every region where it operates, which makes it possible to continuously synchronize applications distributed across multiple facilities for data protection. This, Hamilton said, contrasts with some of the competing services that only provide users with the option of scaling workloads across different geographic areas at the expense of increased latency and potential data loss.

Ultimately, however, the network is only as useful as the other components in the stack. Servers have been a major focus area for Hamilton, who has discovered over the course of his work that Moore’s Law is apparently not the only curve that the market is climbing. “It turns out that mobile tends to be a few years ahead; all the same kinds of innovations that show up there we end up finding in servers a few years later,” he observed, citing power efficiency as one area in which mobile devices are leading the way.

An equally notable dynamic is unfolding in the storage layer, where solid-state memory is removing limitations that have restricted database design for the past few decades to enable unprecedented performance. A testament to that is the Aurora service that Amazon introduced on the second day of re:Invent. It offers five times faster response than a standard MySQL implementation thanks to an architecture specifically constructed to take advantage of flash storage. Moreover, Amazon claims that it does so while maintaining comparable availability with better scalability and security – all for one tenth the cost of traditional alternatives.

Aurora symbolizes everything that the massive multi-tenancy of the public cloud promises to deliver. “Product software is inefficient, when you deliver to thousands of customers you can’t make some of the optimization that we make,” Hamilton said. “Because we run the same thing everywhere, we have a much more reliable product and we’re innovating more quickly. “

Watch the full interview (23:07)