What are serverless computing's most enthusiastic users getting out of the technology, and how are they getting there? They appreciate the ability to implement event-driven architecture, and to support their API deployments. However, they wish serverless had more portability, and would like to have greater local control of features and debugging tools.
These are some of the takeaways from the current Serverless Community Survey, coordinated by the tireless Jeremy Daly, hosted and posted on the GitHub site. In serverless computing, all back-end work such as scaling, capacity planning and maintenance operations is handled in an automated fashion, typically by a public cloud provider, so, in theory, all a developer has to worry about is writing or integrating code for the business problem. Of course, one can argue that the term "serverless" is off, since there is always a server somewhere doing something, but that's another discussion.
By its very nature, this survey is conducted among a self-selected group of serverless proponents, so its focus is on trends and preferences among those already well-ensconced within the serverless world. Accordingly, when asked about the maturity of their serverless efforts, 40% of the 582 respondents indicate their maturity level was "high," that they are "all in on serverless." Another 22% report their embrace is "medium," that they are "transitioning to serverless."
Amazon Web Services emerges as the far-and-away front-runner in this space, cited by 72% of the total sample as their public cloud computing choice. Microsoft Azure follows at 18%, and Google Cloud Platform with 13%. Accordingly, 61% report employing AWS Lambda for Function as a Service, or FaaS, which, along with managed services, form the core of serverless. Another nine percent use Azure Functions. [UPDATE: The figures for the FaaS category are based on percentage of the total base of respondents, including non-respondents. Adjusted for active respondents only for this question (n=391), AWS Lambda comes in with almost the entire subsample at 91%, Azure Functions at 14%, and Google Cloud Functions at 10%.]
The most positive aspect of serverless technology is its ability to enable deployment of event-driven architectures, as cited by 28% of the total base of respondents. Lowered cost of resources to build and support applications follows with 21%, as does the ability to quickly scale applications as needed (21%). The main issue respondents have with the technology is a relative lack of portability, cited by 23%. When asked to write in the features they feel are missing from today's serverless offerings, IT professionals provided a long wish list. The missing features leading the list include best practices, better debugging, cold-start management, greater ease of use, local development, and greater monitoring.
If anyone is wondering if serverless computing can be supported within private clouds, this survey puts to rest any of those lingering thoughts. Serverless is clearly a public cloud play. Close to half of those responding to the question on public versus private, 46%, report that a majority of their production workloads utilize serverless (either through FaaS or managed services) in a public cloud environment, but barely five percent indicate this is the case with on-premises environments. A majority, 73%, report absolutely no serverless workloads even touch their internal infrastructures.
The serverless proponents in this survey are a very busy and prolific bunch. Close to one-third of those responding to the question of volume, 29%, say they now have more than 100 serverless functions in production. The sweet spot, however, is still in the sub-100 range: 26% have between 11 to 50 serverless functions now running in production, and 21% have 50 to 100 instances.
While serverless is being applied to both greenfield and brownfield applications, IT professionals are more inclined to leave their existing legacy applications go for now, the survey also shows. Seventy-five percent answering this question say it is "very likely" their organizations will be building a greenfield serverless application in the next 12 months, versus 34% indicating this is the case for their brownfield applications.
Deploying REST APIs is cited as the most prevalent use case for serverless computing, as seen among close to half of respondents to the survey (47%). Supporting business logic is cited by 33%, as is single-page applications. Another 31% deploy serverless in support of their DevOps initiatives.
Survey data is available for download from the site as an Excel spreadsheet or in CSV format.