Peter Offringa on Edge Compute

I’m well aware that we are weary of more Fastly discussion until earnings, however I think that the following provides an interesting perspective on edge compute applications generally and on architectural issues specifically. Peter Offringa (Software Stack Investing) provided this to me in a recent email exchange after I asked him about the competitive environment in edge compute. He has OK’d me posting it here. I am not a tech guy and I like how Peter manages to make complex (to me anyway) tech issues more digestible. So here is his reply verbatim after I asked him for his big-picture view:

“Edge compute” will enable a broad swath of future technologies. These will range from Internet application delivery, video capture/processing, data streaming, IoT, autonomous vehicles, manufacturing, transportation, etc. Many large technology providers will want to address aspects of this landscape, including the large cloud vendors (Amazon, MS, Google), Nvidia, component manufacturers, embedded systems, etc.

Where I think Fastly (and Cloudflare) will have plenty of room to grow is in the architectural shift in the delivery of Internet-based software applications. These appear to be evolving in three fundamental ways:

1. Towards serverless. Serverless eliminates the need for Development teams to even worry about provisioning servers in order to run their applications. They just upload code to a serverless environment and the hosting provider handles the rest. This is the next logical step in the progression from self-managed hosting (data centers) to cloud-managed hosting (AWS) to fully managed runtimes (serverless). The last big hurdle with serverless for true web application delivery was the cold start time. Now that these times are fast enough to be imperceptible (under 1ms for Fastly and 3-5ms for Cloudflare), serverless can be used for synchronous workloads (when a human is waiting on the response) like a web or mobile app.

2. Away from origin towards the edge of the network. Processing code at the edge just makes the application faster. It also addresses data sovereignty issues by keeping all data local to a country per PoP. With 5g, users will expect more responsive application performance and content delivery. This will be facilitated by running much of the application code in PoPs closest to the user. Making applications run in a distributed fashion (in parallel across many PoPs) is a difficult computer science problem and requires a whole new paradigm.

3. Security. Security is evolving away from the endpoint and back to the network. Vulnerabilities have traditionally been mitigated on devices by installing monitoring agents on them to look for malicious behavior. However, if an enterprise passes all their traffic through an edge network (like Cloudflare One) and the edge network provider has sufficient intrusion detection capabilities (traffic shaping and inspection), then the need for endpoint monitoring diminishes for enterprise assets (servers, phone, laptops, etc.). Similarly, if all traffic for a large publicly facing Internet application is routed through an edge network, then a combination of network and runtime-based application security may suffice (like what Signal Sciences and Cloudflare provide in WAF/RASP). Also, as application hosting moves to serverless, at least for the entry points, then the container “endpoints” become amorphous.

Cloudflare has already evolved along these lines. For Fastly, I view CDN as just the starting point and will follow a similar path, but with more focus on application delivery and security, versus enterprise. CDN is the low-hanging fruit for a software-driven network of global PoPs, and provided a suitable proof-of-concept for Fastly. With Compute@Edge, Fastly will have a platform on which to build a whole slew of new products that are optimized to deliver distributed, serverless solutions from the edge for all kinds of use cases. This job description provides a teaser:
but there would be much more to come, I think. Cloudflare is taking a similar route with Workers and posits that all new products (like Cloudflare One) are built on top of Workers and their distributed network of PoPs.

How the cloud giants react to this will be interesting. I agree with other analysts that they won’t stand by and do nothing. But, I also think they have a lot invested in the status quo and technology evolution often sees the disrupter eventually getting disrupted.

Whether Fastly or Cloudflare will win over the long term remains to be seen, but I think they are both demonstrating enough focus and innovation in their respective areas that it is worth being invested at some level or at least watching closely (depending on your appetite for risk and the obvious bumpiness).