5 Interesting Findings About Serverless
Over two and a half years ago at re:Invent Amazon announced AWS Lambda, a FaaS platform allowing developers to run ephemeral, event-driven code without provisioning or managing servers. Defining a new technology category, Lambda helped usher in a wave of serverless solutions that continue to gain momentum. As serverless inches up the hype cycle, chatter about its future potential has obfuscated the true state of the industry. Our conversations with startups, value-added resellers (VARs), and end users have identified five key findings.
FaaS adoption has been slower than containers. Containers spread like wildfire through the developer community. Docker originally launched in March 2013, and according to DataDog, within two and a half years ~15% of DataDog customers had “adopted” or were “dabbling” with the technology. Over the same duration, our discussions suggest <10% of customers have tinkered with serverless. Unlike shifting from VMs to containers, in which monoliths are decomposed into often existing processes that could be wrapped in a container, FaaS deployments require the developer to further breakdown the process into event-driven single-purpose functions, a higher burden. FaaS is also constrained to stateless components of an application that require less than five minutes of execution duration in Lambda and nine minutes in Google Functions. Finally, each platform has different parameters so functions must be tailored to each cloud’s specific backend services.
It’s all about the $$$. While FaaS platform providers tout the benefits of increased business agility, decreased lead time, and focus on business logic, customers are primarily adopting FaaS for the cost saving. Lambda pricing for each execution is based on memory consumption and execution time. Customers have moved from buying an entire server to hourly billing on EC2 to 100 milliseconds of use on Lambda. Shifting from EC2 to Lambda is 36,000 times more accurate and especially useful for inconsistent loads. As IT budgets remain flat, teams continue to look for cost efficient options like Lambda.
Serverless isn’t here, but it’s coming. VARs building serverless environments for start-ups to enterprise customers noted that conversations have shifted over the past three months. Instead of discussing the movement to containers, customers have become interested in understanding the transition to serverless and are beginning to dip their toe in. Most VARs believe wide spread FaaS adoption will hit in 2–3 years.
Vendor lock-in remains a top concern. Many believe serverless is the ultimate vendor lock-in since third parties provide managed services from the server to the runtime layer. Customers pause because they continue to want visibility and control over their stack. Currently, decisions around which FaaS platform to use are mainly determined by pre-existing relationships with vendors. Over time we believe FaaS platforms will reach feature parity so customers will not be reliant on one vendor for particular functionality. We can image customers toggling between clouds by leveraging third party serverless engines and API gateways so they can play on arbitrage opportunities, decreasing the public cloud vendors’ rents.
Parse may be gone but its scar tissue remains. In April 2013 Facebook acquired Parse, a BaaS that manages data services on the developer’s behalf. By January 2017 Facebook wound down Parse and open sourced the Parse server. Due to the challenges of migrating from a backend platform and fears a startup could go out of business or be shutdown, new vendors focused on the storage or database components of BaaS may experience adoption challenges.
While serverless is a fledging market with many users leveraging the solution to cut costs, we see an increased shift towards these solutions especially for naturally event-driven use cases like scheduled services (e.g. cron jobs), data processing and manipulation, IoT, and social networking (e.g. bots and image manipulation). We expect customers to become less wedded to specific vendors over time and believe BaaS solutions offering a service to replace a repetitive task to have an opportunity.