Wanted: sub-$10 *networked* boards

Morgaine Dinova wrote:

networking is not directly supported in the otherwise-awesome board's software,

That's probably the single biggest barrier.  While there are various network stacks available for various of these class of devices, they all tend to have one problem or another. Single vendor and/or not well integrated forcing you to rethink your software to live with them are not particularly helpful. Usefully supporting Linux (or even getting it to run) on sub $10 hardware is going to be challenging, and as we've seen with the Pi/BBB if you do then you'll just have a hoard of XBMC junkies complaining it doesn't have a decent GPU

Some middle ground is needed here, something with enough of a bsdsockets api to be both useful and familiar territory for the application developers would be a good start.

At least STM32* and Kinetis devices that have onboard ethernet mac are available, but once you add a phy and magnetics I suspect you may have blown the $10 budget. High volume could help with that, but I wonder if the market is ready for something like this. Pre-defined IoT devices that just work, out of the box could probably do ok - the lesson to be learned from the RPi/BBB being that if you need the user to do something simple like loading the software onto a SDCard, then you're probably going to fail.

Morgaine Dinova wrote:

 

Yes I know, it's quite easy to integrate them into an IP network, either through USB-based NICs or wifi adapters or using a small I2C-connected Ethernet or wifi board, or even Bluetooth or NFC, or even using SLIP over RS232 from days gone by.

I'd even go as far as suggesting that something like a usb wifi adapter is going to be a problem on a $10 board, simply because you need to implement TCP/IP and USB and something like wpa_supplicant just to get connected, bluetooth is probably worse. Newer technologies have added lots of layers of complexity you likely don't want in a $10 board, SLIP is better as its quite simple and you only need one out of three of the complex pieces.

Not in any way picking on Olimex here, but this example of an STM* based board with ethernet is nowhere near the $10 segment http://uk.farnell.com/olimex/stm32-e407/stm32f407zgt6-ethernet-usb-dev/dp/2308254

True enough that volume is the enemy, but I'm not seeing any similar boards with ethernet anywhere near $10.  An RPi or BBB or OLinuXino-LIME costs less and is much more capable, so there's another mindset change required to get a $10 board that people will buy in sufficient volume to be able to meet the price point.

Let me know when they appear, I'll buy a few

Note that I very specifically did not mention application processors nor Linux.  I didn't even imply it, since all three "awesome" families use microntroller SoCs,  I am a total fan of the Unix model of computing, but it doesn't have to be reached in one step, and baby steps can make the impossible seem more possible ...

The first step is sub-$10 networked microcontroller boards.  We're extremely close to that already, because there are a few microcontroller SoCs that provide not only on-chip Ethernet MAC but also on-chip PHY.  There isn't really a cost barrier either --- copper coils and ferrite don't add an intrinsic quantum leap in BOM cost, it's mostly marketing.

Once we're there with microcontrollers, application processors have a launching pad.  It won't take long.

Morgaine.

Morgaine Dinova wrote:

 

Note that I very specifically did not mention application processors nor Linux.

Oh yeah, I fully understand that's not what you're looking for. I just think it's crazy that the application processor boards are, currently, cheaper than the microcontroller versions that have ethernet. While that persists there's no market. Like the Pi, all it takes is one to make the jump to a $10 board with ethernet and we'll soon see more.

Of course getting to $10 is going to really upset the people who produce devices like the Lantronix XPort since you'll be significantly cheaper for essentially the same thing.

selsinork wrote:

 

I just think it's crazy that the application processor boards are, currently, cheaper than the microcontroller versions that have ethernet.

Yup.  Nuts, huh?

Morgaine.

I mentioned Internet of Things, but perhaps I should have mentioned how this fits into low-cost networking.

It's simple:  sending a sensor reading every now and then over IP doesn't stress 10Mbps Ethernet.  Not long ago we were doing everything over 56Kbps modems (and 64Kbps ISDN was considered industrial strength!) and thinking how awesomely responsive it was.

So the speed bullet points don't matter so much, compared to the key enabling fact that IP networking is present out of the box.  And that doesn't have to cost a lot of money --- at the end of the day, it just adds some glorified shift registers and 50-year old passive components.  Where there's a will to engineer for low cost, this can be squeezed far below the $10 point including profit.

Morgaine.

USB hardware is very cheap (no magnetics, low power) and usually present all but the cheapest SoCs, so why not do IP over USB directly without messing around with Ethernet?  With 10baseT and higher you need an Ethernet hub or switch anyway, and you could fairly easily make an IPoUSB bridge/router that has a bunch of USB host ports to connect to your devices plus an Ethernet up-link.  The b/router would perform the polling needed to see if devices have upstream IP packets.

I thought something like this might be a nice way to network a classroom filled with RasPi Model A boards.

I've heard lots of talk about the Internet of Digital Things (IDioT), but I haven't seen standards or chips for handling the 'torking (my first neologism for 2014, an abbreviation of "networking" based on its common Internet mis-spelling "newtorking").

John Beetem wrote:

 

why not do IP over USB directly

That sentence carries its own answer, by proposing an impossibility.

The serious answer though is that I dealt with (and eliminated) adding external USB-based NICs and Wifi adapters in my opening post.  My reason first and foremost is that if you have to add networking (over USB or through any other interface) then by definition it's no longer integrated, always present, running out of the box, expected by on-board software, and so on --- in other words, the board is not a networked device by design intent.  It's this design intent and matter of principle that I'm seeking --- that everything that isn't by design intent meant to be isolated should be networked, and networking in 2014 means IP.

Specifically, a board with USB is not a networked board.  You can turn it into one, but that's a different kettle of fish.  Pi Model B is networked, but Pi Model A is not --- it's merely networkable.  Note also that this thread is about microcontroller boards because application processor boards are nowhere close to $10 at this stage of development.  We need to walk before we can run so I'm focusing on $10 networked microcontrollers initially as a baby step.

There's an important secondary problem too:  when they have no integrated networking, microcontroller boards typically do not contain networking software built in either, so even if one accepted a not-out-of-the-box approach and added an external USB NIC, that would get you nowhere by itself.  It becomes a project in its own right to get them networked, instead of it being a default and well supported function.

Note that I haven't eliminated integrated Ethernet or Wifi NICs sitting on a board's USB, in the manner employed on Pi Model B.  Although I agree that it's an option, I don't like that approach because of its high power consumption and poor performance (on Pi at least).  Worst of all though, and really the kiss of death for it, is that it requires quite a costly component.  Since this thread is about the sub-$10 market niche, integrating a USB-NIC or USB-Wifi device is going to be seriously difficult to achieve on cost grounds.  Even external PHY chips are problematic owing to cost, so a SoC providing both Ethernet MAC and PHY (like the old Stellaris Cortex-M3 and others) is really the way to go in this bottom-end niche.

And a Happy New Year 2014 to you and all !!!

Morgaine.

Regarding Internet of Things, stripping off all the frills and hype and decoration (we're pretty immune to hype anyway here, fingers crossed), it's just IPv6 on everything at extremely low cost through integration.  Viewed that way, IoT documentation is already 100%.  (IoT organizations typically push IoT middleware, but that's actually just decoration around what is no more than plain IP.)

Note: IoT on IPv4 is just a wave to the past so that the poor old 98% of IP users don't feel too disenfranchised.

Note that my $10 goal doesn't come even close to embracing IoT in a proper way.  The aim should be sub-$1 if networking is to be omnipresent in everything that isn't by design isolated permanently.  (There are many extremely important applications where such isolation by design is important, indeed crucial, particularly in this post-Snowden era).  Sub-$1 IoT devices are perfectly possible even today, given that many microcontroller families have members that cost a fraction of a dollar.  But for 2014, I'm happy to aim towards just the relatively unambitious $10 niche.

So, don't get hung up by IoT vendor machinations and nods to the past.  It's just highly integrated IPv6.

Morgaine.

Morgaine Dinova wrote:

 

Regarding Internet of Things, stripping off all the frills and hype and decoration (we're pretty immune to hype anyway here, fingers crossed), it's just IPv6 on everything at extremely low cost through integration.

Actually, I was thinking about IDioT layer 1, where there's been talk of ultra-low-power wireless networking, so that you can have IDioT devices that have very long battery lifetimes or preferably do energy harvesting (the 21st Century self-winding watch) so that they don't need batteries at all.

Of course, if we're limiting networking only to devices that I personally want networked, then I'm already done!  I absolutely don't want my TV, my car, my bicycle, my furnace, my cats, my glasses, or my wristwatch networked to anything.