Amazon EC2 Spot Instances Help Autodesk Spend Less, Scale More

As little as a decade ago, most Autodesk software required desktop computers. A user would create a file and render it locally. Renderings were slow, often taking days to finish and locking the computer until the rendering completed. As one of its first forays into software as a service (SaaS), Autodesk used on-premises data centers for rendering. This accelerated rendering, yet the process was resource- and time-intensive, requiring equipment maintenance, space, and capital expense to keep pace with demand. As an early lesson for what cloud computing could offer its customers, RaaS also helped Autodesk see the value in offloading compute-intensive processes to a partner. This opportunity drew Autodesk to AWS and began the long partnership. Today, Autodesk is an Advanced Technology Partner in the AWS Partner Network (APN). “As part of APN, Autodesk has been well positioned to take advantage of innovative cloud services. It created a solid foundation for our own cloud journey and helped us to expand what we could offer our customers,” says Susanna Holt, director of engineering for Autodesk.

As early as 2010, Autodesk explored providing cloud services to its users, beginning with Amazon Elastic Compute Cloud (Amazon EC2) to deliver secure, resizable compute capacity in the cloud. Those instances were always on and available on demand. As user demand swelled, Autodesk simply added more instances. In 2014, Autodesk became an early adopter of Amazon EC2 Spot Instances to manage EC2 costs. Amazon EC2 Spot Instances help users purchase unused EC2 instances at a discount. From then until 2016, Autodesk and AWS worked closely to optimize the use of Amazon EC2 Spot Instances, reducing Autodesk’s instance costs by 50 percent.

Even with cost optimization efforts, in 2017, Autodesk realized its RaaS workload far exceeded other workload costs and again turned to AWS. “Our main goal was to solve our cloud rendering challenges,” notes Holt. “We felt genuine support from the AWS teams during the entire process.”

As Autodesk focused on cost containment, AWS focused on enhancing Spot features by shifting from bid-driven to market-driven pricing and giving users the option to hibernate instances if EC2 capacity was no longer available. When users choose to hibernate, instances resume from their prior state once capacity is again available. This ability means instances resume faster, and workloads experience less disruption.

Autodesk also uses Amazon EC2 Spot Fleet, a collection of Spot Instances. A Spot Fleet starts by creating a request that includes the target capacity based on application needs, the launch specifications for the instances, and the maximum price an organization will pay. Spot Fleet then attempts to launch instances needed to meet the target capacity. Finally, Spot Fleet fulfills the request when the current Spot Instance price is lower than the maximum price specified, and capacity is available. With these changes and added services, the Autodesk RaaS team cut rendering costs in half even while doubling its monthly job count.

“Any company with compute-intensive services should try Amazon EC2 Spot Instances for the best combination of high performance and low cost,” says Xiaoqing Zhuang, software development manager for cloud rendering at Autodesk.

Autodesk falls into the classic use case for Spot Instances. The company has a controlled workflow that can batch jobs and adapt to interruptions when available EC2 capacity falls. Other enterprise operations can take advantage of cost savings from Spot Instances as well, particularly if they support fault-tolerant workloads. Big-data workloads, continuous improvement/continuous deployment (CI/CD) functions, and simulations match well with the benefits provided by Spot Instances.

Zhuang sums it up well, “For companies with services or processes that can run when there is less demand, Spot Instances serve as a terrific way to lower costs. Those workload requests can be cached and run when the Spot Instances market price is most attractive.”

“Our success with RaaS has generated interest from other parts of Autodesk. We’re building on our SaaS experience and planning how we can take that further,” says Holt.

Rather than focusing on costs, the RaaS team now concentrates on more strategic concerns. “The range of services available from AWS freed us to focus on our business logic rather than operations. AWS helps us to deliver both performance and reliability,” says Zhuang.

Spot Instances are having a positive impact on product development, too. “Our good experience with Spot now means that every new service being considered by Autodesk is run through that filter. We’re constantly asking: ‘How can Spot help us be more cost- and time-efficient?’” says Holt.