Function as a Service (FaaS), the popular incarnation of serverless computing, along with containers, has played an important role in the migration from monolithic application architectures to those based on microservices. Its introduction has changed the way developers think about infrastructure and application development. While at the same time, it has also driven evolution in the consumption-based infrastructure models we saw introduced with the advent of elastic computing.
FaaS has simplified broad adoption of value-add platform services from the major cloud service providers (CSPs) such as database and authentication services. It has quickly grown to become a viable infrastructure option for developers and IT professionals alike, serving as another tool in the chest – perfect for short-lived or ephemeral tasks. In recent months, we’ve seen announcements from content delivery network (CDN) providers touting a new breed of FaaS offerings promising low latency, workflow customization and cost containment.
CSP FaaS vs. CDN FaaS
CSP FaaS resides in regional hyper scale data centers with the number of data centers in a region typically in the low single digits. These are the same data centers where the myriad of infrastructure and platform services are offered. Enterprises spin up their applications in virtual private clouds in a single region, sometimes they configure a backup region but even with today’s tools it is still complicated to manage data replication/consistency and ensure environments are kept in lockstep.
In contrast to CSP FaaS, CDN FaaS solutions are highly distributed with various PoP locations. These CDN FaaS architectures are typically integrated with the providers’ edge delivery locations. Depending on vendor architecture, there could be hundreds of edge locations. Whereas CSP FaaS benefits from being collocated with platform services, CDN FaaS benefits from being closest to CDN action, where end-user requests/responses are received and processed. This is where edge computing comes into play because it minimizes latency while allowing for custom content delivery workflows.
Empowering Innovation by Enabling Customization
Developers use functions to customize content delivery workflows to create more compelling user experiences. Up to this point, CDNs provided limited capability for developers to define custom workflows via their own code. Flexibility has been limited to predefined “rewrite rules” and custom behaviors are the product of professional services-led engagements, which is not exactly developer friendly. CDN FaaS changed this, handing power to developers to deploy code and create unique, differentiated user experiences.
The Make Or Break Element: Latency
It is important to minimize latency between application services that rely on FaaS to avoid program timeouts. CSP FaaS options typically offer a broad range of event-triggers and run-times/programming languages to support developer preferences and align with the other elements of the development environment.
CDN FaaS has request triggers that are integrated with end-user or origin server request/response flows. FaaS invocations are typically a result of an end-user requesting content or an origin server providing a response that needs to be customized on a per end-user basis to reflect their preferences. The scale and latency expectations are so demanding that the run-time options are often limited to those that provide the lowest memory footprints and highest performance. CDN FaaS can also take advantage of the caching technologies further improving latency to complete functions – especially when remote service or network calls are involved.
How Do the Costs Stack Up?
For CSP FaaS, the pricing model is two dimensional. It is based on the number of requests combined with the allocated memory and the elapsed time of function execution. Some of the CSPs are offering generous free tiers to entice customer adoption – and remember, Function as a Service drives adoption of many of the CSP’s higher value platform services.
When it comes to CDN FaaS, the pricing models are similar, some organizations are charging for requests only but they limit memory foot and the time a given function is allowed to run. However, there is typically a premium to run FaaS in those highly distributed edge locations. Which is no surprise because cost models are higher for distributed/global build outs than centralized models. Also, due to the nature and demands of web and video services, request volumes are likely to be an order of magnitude higher than CSP FaaS requests.
Not All Faas Services Are Created Equal
Function as a Service is an important part of our infrastructure as a service (IaaS) choice but it’s unlikely enterprises make a CSP solely on the FaaS offering. Typically, the decision is based on what value-add platform servers are available to build and support the application, and how seamless can those services be integrated with FaaS.
For CSP FaaS, it is all about the bigger picture, the infrastructure and platform choices. This is also true with CDN FaaS because it is unlikely a CDN purchasing decision will be made on that CDN’s FaaS capabilities in isolation. The selection criteria will include FaaS but the larger weighting factors will include the content delivery capabilities, such as global reach, capacity, origin offload, feature support and so on.
Overall, CSP FaaS are optimized for flexibility, multiple run-times, broad range of request triggers and tight coupling with CSPs’ higher-value platform services. Latency is important but the focus is within the CSP region or customer’s VPC and the use cases are often enterprise applications for B2B technology companies. Meanwhile CDN FaaS is all about performance, run-time efficiency and latency. It’s about running the functions in conjunction with the CDN – closest to where end-user requests originate and media responses are delivered. CDN FaaS requires enormous scale – on a global scale versus a single, centralized region.
