Live data from Hacker News

Microsoft built its own custom Linux kernel for its new IoT service

techcrunch.com

91–100 of 304 posts

Re: Microsoft built its own custom Linux kernel for its new IoT service

#91

After seeing this I think the Amazon AWS/FreeRTOS combo is looking like the better architecture for a node solution. FreeRTOS can run on lowly M3s at 120Mhz, like the LPC2478. A 500MHz A7 part is way too much silicon for an IoT node. How much external SDRAM and Flash are connected to that MediaTek core?

FreeRTOS is interesting and useful, but also requires knowledge of the device you are running on as well as implementing drivers. Having done a project involving FreeRTOS, compilation flags need to be customized to the platform one is dealing with.

Re: Microsoft built its own custom Linux kernel for its new IoT service

#92
post #62

Earlier quoted context omitted.

Linux is probably a more compact OS for IoT than Windows, but it is still pretty bloated.

Should be interesting to see if Fuschia gains some mindshare. Linux is "bloated" by LOC standards, but once you target a specific architecture, the LOC drops dramatically. The whole LOC drama train is tied to code that doesn't get compiled when you specify the architecture.

To underline this, the LOC has actually dropped in the next version of the kernel (which has just hit -rc1), primarily because a set of no-longer-used architectures were removed.

Re: Microsoft built its own custom Linux kernel for its new IoT service

#93
post #50

Earlier quoted context omitted.

> I'd have thought that impossible a couple of decades ago. Me too. Good thing Google exists now :)

Ah, see what you did there. But 1998 is pretty close to a "couple of decades". And, I do trust MS more than I trust them now.

1998 is in fact precisely a "couple of decades ago".

Re: Microsoft built its own custom Linux kernel for its new IoT service

#95
post #24
post #5

Earlier quoted context omitted.

I can't monitor you or turn your light switch into a "service" that involves reoccurring payments if we don't use the internet to connect.... /s

And you also can't get a security update when someone figures out how to turn your lightbulbs into a botnet.

The status quo seems to already be devices on the internet that can be updated.... but are still apart of a botnet.

Re: Microsoft built its own custom Linux kernel for its new IoT service

#96
post #62

Earlier quoted context omitted.

Should be interesting to see if Fuschia gains some mindshare. Linux is "bloated" by LOC standards, but once you target a specific architecture, the LOC drops dramatically. The whole LOC drama train is tied to code that doesn't get compiled when you specify the architecture.

Are people who call Linux bloated for IoT-like devices really talking about lines of code? I always thought the main focus was on Linux's architectural layout and how closely tied it is to the x86 memory model & the PC in general. Realtime OSs are currently in vogue because they better match how some IoTs devices work, so there's less abstraction that doesn't apply to that circumstances (and therefore improved perfor…

...how closely tied it is to the x86 memory model & the PC in general.

Is it, though? Linux has had NOMMU (running without an MMU at all) support since the early 2000s, and the atomics / barriers are more based on Alpha's memory model than anything else - several of those primitives just compile out to a compiler barrier or nothing at all on x86.

Most heavyweight things you don't need in an IoT context can be compiled-out completely - I think probably the major bit of infrastructure you can't is support for multiple user IDs?

Re: Microsoft built its own custom Linux kernel for its new IoT service

#97

Am I missing something here? The custom linux kernel part isn't interesting at all - in fact, Microsoft has pretty much admitted they can't scale down Windows. What's interesting to me here is Microsoft is building an IoT solution that allows manufacturers to delegate security to Microsoft instead of having to roll their own.

"Don't roll your own security" has been the marching drum of an entire sector of IoT companies working within the connectivity "slice of the pie."

The general mindset has been that iot has a couple slices: the "thing" (air conditioner vibration sensor), how that thing is connected (Ethernet plugged directly into a smart vibration sensor, or vibration sensor plugged into a connected data-recording device), the transmission/storage of that data (cloud solution? Servers on site? Internet y/n?), Analysis of that data (Microsoft IoT platform? Rolled solution? Now defunct Autodesk iot platform?), and finally the acting upon data collected (chief engineer scheduling repairs/maintenance, project manager ordering new motors, whatever).

So many companies have tried "rolling their own" because they got it working on a raspberry pi or Arduino in a week, then find out their connectivity is not secure at scale (let alone that the solution can't scale at all).

Let specialists specialize. No reason not to let big daddy Microsoft handle the messy bits.

Disclosure, I work at electric imp, thus considering myself firmly in the "secure connection and transmission" slice of the pie.

Disclosure 2, we're partnered with Microsoft for their new IoT push, lol.

Re: Microsoft built its own custom Linux kernel for its new IoT service

#98
post #97

Am I missing something here? The custom linux kernel part isn't interesting at all - in fact, Microsoft has pretty much admitted they can't scale down Windows. What's interesting to me here is Microsoft is building an IoT solution that allows manufacturers to delegate security to Microsoft instead of having to roll their own.

"Don't roll your own security" has been the marching drum of an entire sector of IoT companies working within the connectivity "slice of the pie." The general mindset has been that iot has a couple slices: the "thing" (air conditioner vibration sensor), how that thing is connected (Ethernet plugged directly into a smart vibration sensor, or vibration sensor plugged into a connected data-recording device), the transmi…

...and even then, you're not secure forever. So many times the best security practices have been shown to be insufficient. In fact, I wonder what the scoreboard actually would read, 'roll your own' vs 'best practice'? Maybe not all that different.

Re: Microsoft built its own custom Linux kernel for its new IoT service

#99
post #4

Still cannot understand the IoT mentality. It is actually the CoT Communication of Things. Internet shouldn't be required to turn your lights on or off nor be the device communication backbone in an industrial enviornment. Internet should always be the last resort for communicating because of the distance.

There are use cases for which internet becomes the most effective solution. Mostly "extremely distributed products," like a sensor sold by a company offering analysis and preventative maintenance services alongside the product.

Do the security right and the classic internet of shit botnet type vulnerabilities aren't really an issue (knock on wood...)

Anyway, we've managed to survive pitching internet connectivity instead of local network, and several customers have found use cases for it that I don't think local would be feasible for.

In the end it's not just "technically a LAN would be the most secure option here," I mean that's always the case. It can be more "our partners and solution are offering a so-far-unbroken security model, and connecting via the Internet makes the whole manufacturing, distribution, on-site enrolling, and monitoring cheaper and working out of the box."

Post reply on HN