Case Studies in Cloud Decisions: How 4 Tech Companies Decided to Move Out of the Cloud

Moving out cloud blog 1

Many companies decide to keep quiet about their IT infrastructure decisions. Every now and then a company will announce a move to AWS, Azure or Google Cloud, or mention a data center move, but companies rarely talk about the reasons behind these decisions. This makes it difficult for other companies who are looking for advice and guidance from experienced sources when it comes to their own cloud decisions. While an all-cloud approach can be great for helping new companies get their feet off the ground, many companies end up opting for a hybrid approach to their infrastructure, or owning all their own infrastructure once they have grown. Here are four helpful case studies of tech companies who didn’t keep quiet about their transitions out of the cloud and why they did it.

GitLab

GitLab is an open source software development platform that launched in 2011, and it has grown to service a global community. GitLab consistently and transparently monitors their IT performance, and they have documented the evolution of their infrastructure as they have grown. The company feels a responsibility to document their IT growing pains to help other companies. Pablo Carranza, a production engineer for GitLab, says there isn’t much information out there about how growing companies have tackled infrastructure issues and arrived at their solutions.

The company was having problems with storage and built CephFS clusters to overcome capacity and performance issues with NFS appliances. They were also experiencing outages from high database load. They discovered that running a high performance distributed filesystem on the cloud (CephFS) was problematic, because in the cloud, there’s no minimum SLA for IO latency.

“The cloud is timesharing, i.e. you share the machine with others on the providers resources,” said Carranza on the company blog. “You’re paying a lot of money to get a subpar level of service while still needing more performance.”

GitLab discovered the cloud has a performance threshold, and if they wanted more, they’d have to pay significantly more and deal with latencies, or leave. “If we wanted to make the disk reach something, we would have to wait 100 ms latency. That’s basically telling us to wait 8 years.”

GitHub opted for owning their own infrastructure because, although there were costs upfront and maintenance costs, in the end it would make the company more “efficient, consistent and reliable.”

They believe that the cloud is cheaper and sufficient at a small scale, but not after passing a certain point. They discovered that when adding heaving IOPS, the cloud became ineffective and expensive. For more details about the company’s decision, read their blog post, Why Choose Bare Metal.

Dropbox

Dropbox is a personal cloud and file hosting service for businesses and individuals serving 500 million people around the world. The company always maintained a hybrid infrastructure, but stored all customer file content in Amazon’s cloud. In 2014, the company decided to begin shifting storage out of AWS and onto their own infrastructure.

In a blog post, Dropbox’s VP of Infrastructure, Akhil Gupta, recognizes Amazon as “an invaluable partner” that helped the company grow, and mentions they will continue to use some of Amazon’s services. However, Wired reported that Dropbox gets “substantial economic value” by operating its own infrastructure.

In addition to cutting expenses, Gupta wanted end-to-end control of the infrastructure, and so the company embarked on building their own exabyte-scale storage system with encrypted data at rest and 99.99% availability. By the end of 2015, Dropbox had migrated 90% of their data in-house.

“By optimizing the stack and customizing the infrastructure to our use case, we were able to provide a key differentiator in the market and a key value to our users,” Gupta told Network World. Read more about the Dropbox transition out of the cloud on their blog.

Moz

Moz is a leading SEO company that started in 2004. The company began as an SEO consulting firm and has developed several platforms for digital marketers over the years, including Follwerwonk, Moz Pro and Moz Analytics.

Unlike Dropbox who stated that AWS was a valuable partner, Moz CEO Sarah Bird has been outright critical of AWS: “Over the years, we’ve spent many small fortunes at Amazon Web Services,” stated Bird in a blog post. The vast majority of their total cost of goods sold was for AWS, $7.2 out of $10.8 million to be more precise.

The return to profitability plan for Moz included transitioning away from AWS to their own private cloud. Bird says by owning their own infrastructure, the company is spending less and has improved reliability and efficiency. Moz began moving resources out of the cloud back in 2012. In 2014 the company had returned to profitability and revenue has continued to increase. As of the end of 2016, Moz’s cloud spend was down to $2.85 million. To learn more, read Moz’s annual reports found on their blog.

MemSQL

MemSQL is a leading database platform for real-time analytics that launched in 2013. It is a platform that combines the power of database, data warehouse and streaming workloads in one system. MemSQL is a perfect example of a company acquiring their own infrastructure before their cloud bill spun out of control.

When the company started, renting VMs from AWS was “obviously the way to go,” mentioned co-founder Eric Frenkiel to Wired magazine, “but the bill started to rise.” Frenkiel also believes the cloud is great for businesses that have to manage a dynamic workload, but MemSQL’s workload was fairly consistent.

As reported by Wired, MemSQL was spending $324,000 a year on AWS, but for just $120,000, the company could buy all the physical servers it needed to get the job done. So the company decided to acquire its own infrastructure, and will continue to add more servers over time. “With Amazon, [the company] would have spent about $900,000 over a three-year period. But with physical servers, the cost will be closer to $200,000. The hardware will pay for itself in about four months.”

To read more about companies who chose a hybrid infrastructure or owning all their own infrastructure over the public cloud, download our analyst white paper, Infrastructure Wars: Colocation Vs Cloud.