IoT musical chairs, chiplets, and more

Every Wednesday, Nick Chase and Eric Gregory from Mirantis go over the week’s cloud native and industry news on the Radio Cloud Native podcast.

This week, John Jainschigg stepped in for Nick, and John and Eric discussed:

• Heroku discontinuing free product plans

• Microsoft and Google's game of "musical chairs" in their contrasting approaches to IoT

• Intel's new "chiplet" subassemblies for Meteor Lake CPUs

• And more on the podcast, including phishing attacks on PyPi and Bun's funding announcement for Oven

You can download the podcast from Apple Podcasts, Spotify, or wherever you get your podcasts. If you'd like to tune into the next show live, follow Mirantis on LinkedIn to receive our announcement of the next broadcast.

Eric: Heroku announced last week that they will be ending their free product plans, a move they framed as a “focus on mission critical.” According to a blog post entitled “Heroku’s Next Chapter” by Heroku General Manager and Salesforce EVP Bob Wise:

“Starting October 26, 2022, we will begin deleting inactive accounts and associated storage for accounts that have been inactive for over a year. Starting November 28, 2022, we plan to stop offering free product plans and plan to start shutting down free dynos and data services. We will be sending out a series of email communications to affected users."

The blog post goes on to say that fraud and abuse mitigation on free accounts is consuming an inordinate amount of product, engineering, and security team time. 

In addition to the end of free product plans, Heroku shared a public roadmap on GitHub, inviting public feedback.

The change marks the end of an era, with Heroku’s free tier having really provided an on-ramp and defined the PaaS experience for a lot of beginning and individual developers. Heroku acknowledged this legacy in its blog post, noting that they will be launching dedicated student and non-profit programs in the near future.

John: About two weeks back, Google announced that it was shutting down its IoT core business next year. The emotionally-invested multitudes reminded us that Google has a habit of shutting down services – there was some gnashing of teeth. But at the time, Google’s plan seemed sensible. As one analyst put it, there hasn’t been much innovation or opinionation on IoT services from the major public cloud vendors. And this has created opportunity for third parties specialized and invested in IoT – tiny companies like Bosch and Cisco and vendors like PTC – to create compelling offerings for customers that run on those same public clouds. So Google says they want IoT users, in future, to make deals with these partners, and they’ll just run whatever workloads the partners boost to their infrastructure.

Makes sense, and points up the fact that value in cloud is moving up the stack. If you’re a provider, absolutely worry about how to germinate new markets and ecosystems to get workloads onto your infra, which is how you get paid. If you’re not a provider, climb the value ladder out of infrastructure and into differentiated apps delivered quickly, so you can leverage opportunities as they emerge and crystallize. All open windows are of limited duration.

BUT … Microsoft, given the same business conditions, is making interestingly-different calls. They reorganized their IoT engineering and product management teams in April as part of the Azure Edge + Platform group, and now they’re consolidating IoT and edge teams and building out new IoT-oriented services on Azure. These include hybrid infrastructure services, 5G connected offerings, and their whole family of appliances, plus IoT hubs, mapping services, and OS offerings for devices.

A lot of this seems to pivot on Kubernetes, and on the gradual shift in IoT and edge towards ‘small datacenter’ models, and on ways of using Kubernetes federation and other approaches to unify infrastructure functionality and strategy across core, edge, and IoT devices. Something we’ve been doing at Mirantis for a couple years.

John: Intel announced last week that its upcoming Meteor Lake 14th generation core processors and subsequent Arrow Lake products will be stitched together from multiple “chiplets” – subassemblies combined using its multilayer Foveros packaging tech. Chiplets, or ‘tiles’ are rectangular subassemblies for major chip function groups: the CPU itself, system-on-chip, graphics, I/O, that co-reside adjacent to one another on a base tile and substrate. The goal of the approach is to get the benefits of disaggregation – being able to build chips in a series with different elements, and differently-configured elements (e.g., core counts), but the same architecture, saving the cost of adding features or filling in narrow market requirements while reducing re-engineering. You could like … think about building edge routers this way, for example.

What’s fascinating about this story (to me anyway) is that it enables asynchronous development of chiplets and base architectures, just like containerized microservice-based software development on Kubernetes. During the (longish) lifespan of a base architecture, you can come up a new GPU tile and just drop it in. Same notion as pushing a new and appropriately-signed microservice version container to production, without breaking an app.

Check out the podcast for more of this week's stories.