FittedCloud Uses Machine Learning to Cut Cloud Costs
Over the last ten years, companies of all sizes have enthusiastically made cloud infrastructure a central part of their business operations. But as cloud expenditures have grown as a percentage of the bottom line, an increasing amount of scrutiny has been given to costs associated with services like cloud computing and storage.
In some cases, companies have tasked a single unfortunate cloud-savvy employee with bringing the costs under control, a process that requires hours of analyzing data and tweaking individual cloud provisions—often causing disruption to business operations. In fact, optimizing cloud services is becoming a full-time job for people-given titles like Cloud Administrator or Cloud Manager.
Still, in most business environments, perfectly optimizing cloud provisions in real time has been next to impossible because of the complexity of the problem.
Contrary to popular belief, cloud service providers like Amazon Web Services don’t charge customers based on the exact amount of resources they use, but rather on a number of resources they provision for use.
“Today the burden of making sure consumers provision the right amount of cloud resources is on the customer,” FittedCloud co-founder Prakash Manden explains. “For example, on the computing side, a company might provision a compute instance with eight CPUs and 32 gigabytes of RAM. Cloud providers don’t really care if you utilize those resources or not.”
In some cases, like with cloud computing, service providers can give customers an idea of how much data they’re using. But when it comes to determining block storage and memory utilization customers are largely on their own.
Manden, along with his longtime business colleagues Jin Ren and Jack Huang, observed this problem in 2015 while working with large technology companies at VeloBit, a startup that improves data storage efficiency that they helped build and that was eventually acquired by HGST.
Later in that year, one of VeloBit’s software engineers moved to a new company where he was charged with managing its AWS applications. The company’s spending on cloud services have steadily grown month over month and had reached half a million dollars in the month he finally set out to tackle the problem.
“The difficulty often comes when businesses scale, because it’s very hard to monitor and manage cloud services optimally by yourself, at least without disrupting applications,” Manden explains. “That’s why you’re hearing horror stories about costs because it’s easy to start but it also gets out of control easily.”
The engineer eventually came to the conclusion that the company could save around $250,000 a month, or half of what they were spending, by optimizing cloud provisions. Desperate for help, he looked for services that would automatically adjust provisions based on usage but found nothing.
“That’s when we realized that this is a real problem happening in customer environments,” Manden remembers.
By the end of the year, as the engineer laboriously worked to create a personalized solution for his company, Manden, Ren and Huang launched FittedCloud to solve the same problem for businesses everywhere.
Today FittedCloud’s software uses machine learning algorithms to detect trends and predict customers’ AWS cloud usage. From there the company can either provide actionable insights based on those trends or automatically adjust AWS provisions on its own.
If customers elect for what Manden calls semi-automation, they can view usage reports generated by FittedCloud’s system that show opportunities for provision optimization, ranked based on potential cost savings. Then users can simply click through the advisories and implement the changes in FittedCloud’s UI.
“We provide actions specific to the advisories, that’s the differentiation we have: There are not many companies that provide extensive actions with their advisories,” Manden says.
If customers choose FittedCloud’s full automation solution, users make an account and enter preferences and limits for resource allocation, then the system will take actions conforming to those rules automatically. And although the provisions are based on each company’s past usage, FittedCloud’s models can also adjust in real time to deal with unexpected workload spikes.
“We wanted our solution to work in a manner that’s automatic, transparent and optimum,” Manden says. “Automatic because it’s not practical for users to do this on their own; transparent in the sense that it’s easy to use and there’s no application disruptions, so you never have to shut anything down, and by optimum I mean we use machine learning to provision cloud services in the most accurate and optimized way possible.”
The solution can bring major cost savings to cloud-reliant businesses, which Manden says generally use around 10 to 20% of what’s provisioned to them. Savings can be easily tracked in FittedCloud’s UI.
“A lot of customers are interested in learning how over-provisioned they are,” Manden says. “As part of our advisory we calculate the unit cost of each resource and what the savings would be if they implement our recommendations. All of the details are there for customers, and numbers speak louder than words.”
FittedCloud typically charges two percent of a customer’s overall cloud expenditures, although many customers are still in trials as the FittedCloud team works to add features to the system. Anomaly detection was implemented earlier this month and Manden says more features are on the way.
Beyond 2017, the FittedCloud team hopes to spread its system to cloud services provided by Microsoft, Google and Oracle. For now, though, the team is keeping busy working exclusively in AWS, which had a 31 percent global market share of global cloud infrastructure as of Q1 this year according to market analyst firm Canalys.
Manden says large businesses that are cloud service veterans and medium-sized startups scaling quickly are the demographics struggling with cloud costs the most. When you consider the increasing number of companies investing heavily in cloud infrastructure it would seem FittedCloud’s potential pool of customers is large.
All this indicates that FittedCloud is primed to grow quickly. I hope they’re on the cloud.
Images courtesy of FittedCloud.