Facebook moves Instagram photos out of Amazon and into own datacentre

Facebook has moved all the user photos on Instagram out of Amazon’s cloud platform into its own internal datacentre and cut costs

Facebook has moved all the user photos on Instagram out of Amazon’s cloud platform and into its own internal datacentre. The relocation has cut the social network's costs and reduced its server estate by one-third.

Facebook acquired Instagram in 2012 for about $1bn in cash and stocks. At that time, the deal was the biggest for a profitless start-up since Google bought YouTube for $1.65bn in 2006.

“We bought Instagram two years ago and it was a pretty typical startup – it had about 20 people working on it and its services ran on Amazon Web Services, like most startups do,” said Facebook’s infrastructure engineer, Charlie Manese, in a YouTube interview with Australian datacentre service provider Infrx.

“We have recently moved Instagram photo services onto the Facebook infrastructure and today we use about one-third less servers to service Instagram photos than we did before and obviously we reduced cost from what it was when it was on the Amazon platform.”

The Instagram photo service has about 45 million photo updates every day and 16 billion pictures are shared on the website, said Manese. But this photostore is smaller than Facebook’s photo scale, which has 350 million picture uploads each day and 400 billion photo shares and updates.

Facebook currently has 1.23 billion monthly users and processes 6 billion “likes” a day in real time and more than 200 billion friend connections, he added. “That’s a typical workload of a high-performance compute environment and a big data requirement.”

Manese said smaller companies that are processing big data and providing large-scale web services, such as Instagram, are reaching a tipping point. “A lot of smaller companies are going to start on a managed services environment,” he said. “But there’s a tipping point when it makes sense to move all your infrastructure in-house and we have been able to show and do that.”

Through the Open Compute Project, Facebook shares the lessons it has learnt, the specs and designs of its datacentres, and details of its server equipment. The company currently has four datacentres – one in East Coast US, one in West Coast US, one in Europe – the Arctic Lulea datacentre in Sweden, and its latest one in Iowa, central US.

By 2013, Facebook had reduced its costs by 24% and increased its IT energy-efficiency by 38% since it began using open-source hardware systems in its datacentres, Frank Frankovsky, the social network’s vice-president of hardware design and supply chain operations, had said earlier.

The open-source systems are based on the Open Compute Project, which was initiated in April 2011 by a small group of Facebook engineers looking for ways to scale the company’s computing infrastructure in the most efficient and economical way possible.

Read more on Clustering for high availability and HPC