Food for Thought: Savoring a Feast of Networking Predictions
The impulse is irresistible, even though people invariably know it’s a fool’s errand. Yes, I’m alluding to the annual predictions that pundits and representatives of IT vendors issue every year in late December or early January. In my career as a market analyst, I was often compelled to consult my befogged crystal ball or to read the industry’s gory entrails. I got some things right, but if I’d been clever enough to be right on all counts, I would have been playing the markets fulltime and not working a day job.
Nobody is omniscient. Even the tech gazillionaires, worshipped by legions of sycophants, stumble occasionally, sometimes disastrously. I need not mention names and instances. You know who they are, and when they’ve performed inadvertent illusionist tricks that caused billions of dollars vanish.
One trap that snares prognosticators involves permitting oneself to arbitrarily target a fixed number of predictions before one has considered the nature of the predictions. Let’s say you start with the notion of providing ten predictions for the year ahead. What happens next? The first few predictions leap to mind with alacrity, but when you come to the last few, you suddenly find you’re running short of inspiration. What follows is forecast filler. It’s like one of the pretentious concept albums that progressive rock bands used to foist on an unsuspecting public back in the 1970s: two or three strong tracks interspersed with a lot of perfunctory noodling.
Nico Vibert, senior technical marketing engineer at Isovalent, did not fall into the trap of setting an arbitrary number for his prediction count. When posterity renders its ultimate judgment, I don’t know whether he’ll bat above average on his swings. It doesn’t matter, because his predictions are provocative, well worth reviewing and considering if you have an interest in the future of networking and eBPF, the latter being the foundational technology that Isovalent has leveraged in developing its open-source and commercial tools and products.
As with all vendor predictions, even the best of them, you will detect a self-serving aspect and an occasionally narrow frame of reference to Vibert’s calls, but if you keep those reservations in mind, you’ll still find more substance than bluster in what he provides.
When I was plying my trade at IDC, I was enthusiastic about the prospects of a range of technologies that addressed the needs of what I’ll call, for want of other terminology, the new networking, which was driven by cloud, cloud-native application environments, and the increasing digitization of organizations across industries and geographic regions. One of those technologies was the combination of eBPF and Cilium, which took networking, observability, and network security into a sandbox in the Linux kernel, where event-based policies could automate effective and proximate action to expedite and safeguard application performance.
Many of you will know that Cisco recently acquired Isovalent. It was a far-sighted move by Cisco, perhaps one of the most significant and far-reaching Cisco has made in a long time, but we’ll park that discussion for another time. It deserves its own post. I’ll just mention briefly here that Vibert, who once worked at Cisco, will now return to Cisco. Cue the Welcome Back, Kotter theme music. (That pop-culture reference dates me, as do so many others, but you wouldn’t want me to force unnatural references to brostep (I had to look that up) and the latest series on Netflix, would you? Well, would you?)
Digging In
For now, let’s review Vibert’s predictions, starting with what I believe is his blandest salvo, namely that IPv6-only Kubernetes clusters will become the norm. Look, this prediction might prove accurate, but pundits have been foretelling IPv6 ubiquity for the last decade. At a certain point, the recurrence of the prediction gets a little old.
Eventually, though, it will happen. Still, it’s the longest gestation period in networking since ISDN, which never did meet expectations and was ultimately taken to be an acronym for “it still does nothing.” I don’t believe IPv6 will judder into oblivion like ISDN did, but its eventual predominance will be an anticlimax following its well-established legacy of perennial underachievement.
Other than the de rigueur IPv6 prognostication, the others qualify as savory fare. Vibert begins proceedings with a few eBPF predictions, which seem warranted, even accounting for self-interest, and well within the bounds of reasonable expectations. At first glance, a couple might seem contradictory – the forecast that eBPF will continue to its explosive growth juxtaposed with the opinion that eBPF misuse will result in eBPF fatigue – but I think Vibert does enough to square that ungainly circle, if not wholly reconcile the two perspectives.
There’s a slight bobble where he posits that Kubernetes users will fight back against complexity. He’s right, of course, to suggest that multitudes of Kubernetes users are frustrated with tool sprawl and the overall effect of daunting complexity, but that’s been true now for some time.
Unquestionably, platform and DevOps teams want simplicity, they want complexity abstracted and mitigated, but they’ve been singing that mournful dirge for at least a couple years. In turn, the Kubernetes community has pledged to be attentive and responsive to their complaints. But not enough has been done, and complexity endures. Perhaps we have reached a breaking point, where the Kubernetes ecosystem has no choice but to submit to substantive consolidation and meaningful simplification. I hope it does, but the prospect of instant gratification should be countervailed by a modicum of doubt. Push for a simpler life, of course, but don’t count on it happening overnight.
Vibert’s next prediction is more ambitious. He suggests that projects such as Cilium Mesh will bridge the gap between Kubernetes, VMs, serverless, and bare metal, addressing the needs of heterogenous networking environments. This is a definite requirement, critical for eBPF/Cilium’s growth prospects and necessary to meet the needs of enterprise customers, but it won’t be easy. Still, the prospects are encouraging, and the goal is in sight.
The next prediction, relating to an awkward relationship between platform engineering and networking, hits the bull’s eye. Here’s a key paragraph from Vibert:
I am really looking forward to seeing the intersections of platform engineering and networking. I guarantee you that some platform engineers will be absolutely terrified to let developers have self-service networking power.
Saving the Best for Last
In my view, the challenge of providing developers with latitude and flexibility while avoiding network and security catastrophe will require continuously intelligent automation that verifies and validates choices and configurations from Day 0 through Day N. The platform team will have to provide constant control, but with an invisible touch. The developers won’t know they’re being saved from themselves, even as a guardian angel hovers above them.
After a Web Assembly (Wasm) prediction, Vibert gives us two observability-related offerings. In the first, he cites data from a recent KubeCon where respondents put observability at the top of their topics of interest. Given the application- and service-layer essence of cloud-native environments, observability is more important than ever. The problem is, as the next prediction notes, the burden associated with observability will have to be reduced. The resources consumed by observability are too great in many cases, and only effective tools that lessen the load should gain favor.
Unlike those who scramble and scuffle to round out their last few prognostications, Vibert serves his best predictions for the end of the feast.
Among them is his claim that the networking industry is “bracing for change.” His choice of verb is apt, for the change coming to the industry won’t necessarily be universally welcome.
At this point, Vibert plays his expository cards well, first contending that Broadcom’s acquisition of VMware will cause some organizations to reconsider their commitment to the VMware technology stack, including their attachment to NSX for network virtualization. I agree. Broadcom is the new Oracle, and not in a good way, especially if you’re a customer paying dearly for the privilege of being locked into a gilded cage.
Vibert then points to the growing popularity and effectiveness of open-source networking. This was inevitable, I believe, as the perceived and real value of networking moved up the stack in deference to the applications and services on which organizations depend. As networking and software commingle, the former follows the same rules as the latter, moving away from hardware-defined standards bodies and more toward community-centered software.
The shift is reinforced as networking has become yoked to Linux, which hosts (pardon the pun) a community comprising a cornucopia of open-source projects, a growing number of which are dedicated to various facets of networking. There’s also the generational dimension, which I have mentioned previously, whereby younger platform engineers are supplanting the siloed and increasingly superannuated masters of complexity who formerly ruled the networking domain.
Profound change has arrived in the working world, and for some struggling with the transition from the old to the new, a lyrical snippet from Jarvis Cocker (of Pulp) might seem apt: “You’re gonna like it, but not a lot.” Vibert’s use of “bracing” is about right, even if change was inevitable.
Of perhaps greatest interest are Vibert’s two predictions relating to the confluence of AI/LLMs and networking, a busy intersection where collisions are sure to occur. First, Vibert foresees that “networking tools will begin to natively integrate with LLMs to provide a chat-like experience to interact with the network.”
He provides some examples of what might become possible:
For example – imagine being able to ask why your users complained last night and being told by a chatbot that traffic dropped between 10 and 11PM and that the likely culprit was a misconfigured network policy.
I’d love to give simple instructions to a chatbot about the architecture I want to build and in response, to get a ready-made Terraform configuration.
I’d love to talk to my network and ask it why it’s stuck in a routing loop.
The reader might at this point be nodding enthusiastically in agreement, but there’s a catch, which arrives in Vibert’s subsequent prediction regarding AI and networking: Using LLMs to troubleshoot network issues and generate network configurations could end in despair. Indeed, the unwary could be hoisted on their own automated petards. I agree with his admonition that, if you’re going to use ChatGPT or other genAI LLMs for network-automation purposes, you must first ensure that you thoroughly understand the workings of the machinery under the hood. Mystery lurks in the unknown, and surprises are rarely welcome on a production network.
These are good predictions, valuable not so much as potentially ego-gratifying boasts but because they encourage us to think meaningfully about what the future might bring.