⬅️ When We Was Chip Fab Focused 🧭 Busy Child ➡️

Get It All While You Can

July 25, 2022 by Jay Cuthrell newsletter cloud repatriation

Tesla - Ez Come Ez Go (1986)

This week I wanted to thank readers for all the feedback and sharing. Being able to maintain my publishing streak for over a year has been incredibly rewarding.

Thanks for reading Fudge Sunday! Subscribe for free to receive new posts and support my work.

Special note: Thanks @VirtSecurity Jonathan Copeland for review of early drafts

Getting Informed

Several readers have asked to revisit topics from prior newsletter issues from a year ago. One such topic is the reality of repatriation from edge to core to cloud and I’ve decided to industrialize this concept as continuous commensurate repatriation.

Let’s get into it.

Share Fudge Sunday

Sometimes it just feels so right 🎶

First, it’s worth looking back at the notion of edge to core to cloud that is blurring which I covered this time last year. To recap:

And other times so wrong 🎶

Second, the topic of moving from edge to core to cloud for workloads sometimes feels similar to the “Riddle of the Sphinx”1 but with security and finance concerns being the limiters to finding an answer. Indeed, sometimes the “xOps solves this” meme answers invite the Sphinx to feast upon pundit travelers.

So, again, let’s look back at a few recent issues for guidance from March 2022:

Also, consider recent posts from Grey Meyer2, Ed Sim3, and Tom Tunguz4:

The change goes on and on and on and on 🎶

Third, if there is a multiverse for multicloud, then perhaps there is an applicable Spider-Man: No Way Home (2021) meme. But, it’s important we be careful about click driven punditry and tech journalism that seeks to make everything into an overly simplistic zero-sum game.5

So, (spoiler alert?) perhaps we’re all on a path akin to residual self-confirming equilibrium that academics like Yakov Babichenko or Aviad Rubinstein may generalize one day6.

In effect, there are so many R’s (beyond the normal 6) to consider as I outlined in a newsletter issue from last year:7

a superfecta of R’s at the minimum seems more likely instead of assuming everything goes to the public cloud. Retain Relegate Renovate Resuscitate Retire Replatform Repurchase Refactor Rehost Repatriate Refinance Reinvest Rationalize Recombine (etc…)

When what we think of as data centers are becoming micro data centers as invisible or nondescript base pedestals in neighborhoods, unmarked rooms in unmarked floors in unremarkable buildings, and floating overhead a la Starlink / Kuiper /etc. – it’s gonna get funky. So, it’s reasonable to expect four more R’s might be Relative, Reallocate, Readied, Realistic.

To gather just a few examples that many readers might recognize, the VxRail, Nutanix, and Scale Computing approaches all proved the need for discreet clusters for smaller and smaller things in data centers. Now, as ARM, discrete in situ cooling8, and increasingly commensurate placement of workloads takes hold (again?) it’s gonna be fascinating.

Precision immersion cooling from Iceotope at HPE Discover 2022So, on a long enough timeline, if there are k8s/k3s clusters running in your local Chick-fil-A then it’s incredibly likely that evolved approaches are going to eventually filter out to other firms and plate / power / ping modalities.9 Further, the economics of deterministic power, weight, cooling, and geometry in data center footprint are only going to become ever finer grained year over year.