Meet Linux's Little Brother Zephyr, a Tiny Open Source IoT RTOS (linuxgizmos.com) 130
DeviceGuru writes: The Linux Foundation has launched the Zephyr Project, to foster an open source, small footprint, modular, scalable, connected, real-time OS for IoT devices. The Zephyr Project's RTOS implements both a small footpoint microkernel and an even tinier nanokernel, and is the result of Wind River contributing its Rocket RTOS kernel to the Zephyr Project. (Wind's Rocket RTOS will now become a downstream commercial distribution based on Zephyr sources.) To get a sense of Zephyr's benefit, its nanokernel is said to be able to run in as little as 10KB of RAM on 32-bit microcontrollers, whereas a minimalistic Linux implementation like uClinux needs upwards of 200KB. The Linux Foundation hopes to see cross-project collaboration between the Zephyr and Linux communities. Technical details are at the Zephyr site.
IoT (Score:2)
So, it's a tiny computer. What makes a device an IoT device? Does it just need to be a network connected device? Is a Raspberry Pi an IoT device? Is a smart phone? An IP phone? A smart TV? A desktop computer? A laptop? A netbook? A tablet? I need a solid definition so I wont be annoyed when I see "IoT".
Re:IoT (Score:5, Informative)
I guess the definition for IoT devices is something that is "a device not primarily used for computation that can be controlled/accessed through the Internet". For example a washing machine, a watch, a fridge, heat controllers, windows (the physical ones that you can open), smart meters.
IoT then also refers to a time where these would become extremely commonplace, in the sense that many types of devices are connected to the internet in everyones home/work/public areas.
Re:IoT (Score:5, Funny)
What makes a device an IoT device?
If it floats on a cloud, runs off the willpower of app developers and is a security and privacy nightmare. Then it's IoT.
Re: (Score:1)
The account https://twitter.com/internetofshit [twitter.com] is a gold one!
Re: (Score:1)
Re: (Score:2)
MS *are* targetting RPi2 as a Win 10 IoT platform.
(Basically a phone without the phone part, which some alluded to as fulfilling the definition)
Re: (Score:1)
Nope, we're concerned about security on the devices we make. So when I hear 10K RAM my first thought is that is that it chopped off security from the feature list.
Most of these IoT related operating systems, of which there are many with that claim, tend to either be very low on features, are meant for larger systems than are practical for long term battery power, or are very commonly unfinished and barely out of the proposal stage. It is not uncommon for a group of big name companies to announce they're f
Re: (Score:2)
I head that in Jeff Foxworthy's voice.
Re: (Score:3, Insightful)
Don't know why some idiot modded you down. It's a legitimate question.
IMHO it's pretty hard to define IoT in any way that doesn't include smartphones, unless the definition contains words like, "..but isn't a phone."
Or maybe it's cost. If it costs over $50, then it's not a thing: it almost certainly has a more specific name, like "phone" or "tablet" so that it doesn't have to fall back to using its general class of "thing."
(I used to think it might be defined by whether or not the device does any I/O direct
Re: (Score:2)
Or an id short of an idiot.
Re: (Score:3)
IoT refers to an object that has no really good reason to be connected to the Internet, and a few reasons why it shouldn't, but will be anyhow. This is all in an attempt to use up all of the IPv6 IP address space.
Re: (Score:2)
More to the point, Zephyr is a new RTOS that's just like FreeRTOS, RTEMS, eCOS, EmbOS, uITRON, ThreadX, VxWorks (shudder), and others, but without the significant support ecosystem and established pedigree.
Just one question: Why?
Re: (Score:2)
I wish someone could define it in a highly specific way rather than just using it as a umbrella term for so many distinct devices; desktop PCs, laptops, ultrabooks, Macs, calculators, phones, abaci, difference engines etc.
"IoT" means Internet of Things. So like a "computer" is anything that can compute, "IoT" as a thing that has some connectivity to the Internet. So yes, all of the things you mentioned could fit into the definition of "IoT". Smart TV = probably is a thing with Intern
Re: (Score:2)
If it's a thing, and it connects to the Internet, it's part of the Internet of Things.
Usually this is reserved for "things" whose primary purpose is not what you'd use a desktop computer for. A RasPi hooked up to a keyboard and monitor running LibreOffice? Not a thing. The same RasPi running headless to control your hot tub? Thing.
It's the same blurry line that differentiates "embedded" and "non-embedded" systems. On one side you have devices that are clearly embedded, IoT devices. On the other side
OS for the new "Microcontroller" (Score:4, Informative)
I've been seeing piles and piles of new chips that come with integrated 32bit arm cores, flash, and ram. And they're dirt cheap. Nowadays it seems pretty much all non-trivial chips include an ARM core with enough storage and ram to run circles around any desktop computer found in the 80s or even early 90s.
About 9 dollars shipped for a hacker-friendly breakout board based on the NRF51822.. And that's not just a processor. It's a Bluetooth 4.0 radio that supports bluetooth LE and all the other bells and whistles. 15kb ram, 256k flash, and whole mess of features to get data in and out. Seems right in the footprint for this little OS.
Add bluetooth and a fairly powerful "micro" to any project for less than 10 bucks.
http://www.ebay.com/itm/-/400970594571
Development board for the above module:
http://www.ebay.com/itm/-/400967553827
Browse that seller of the above items and be amazed at the absolutely insane amount of dirt cheap hacker friendly electronics you can get shipped from china for just a few bucks. Granted specs may be exaggerated and you'll want to do a little poking and testing on your own.. But that's half the fun, isnt it? :)
Re: (Score:1)
electronics you can get shipped from china for just a few bucks
I'd be willing to pay five times the asked price for items like these if they were manufactured in Europe.
Re:OS for the new "Microcontroller" (Score:5, Informative)
Raspberry Pi's are mode in Wales.
Olimux devices are made in Bulgaria.
Re: (Score:3)
And SoCs aren't even really made in China, are they? China isn't a big player in semiconductor manufacturing unless one counts Taiwan as China.
China assembles a lot of electronic stuff but I don't believe they are a big player in actually manufacturing it. USA/South Korea/Europe/Japan/Taiwan are.
Global economy and all...
Re:OS for the new "Microcontroller" (Score:4, Informative)
The issue is that the cheap gizmos are the weak link in the chain. That, and the "hidden third shift". In Europe, if I spec something, I will get what is specced, every time, from now, to whenever production stops. I have read over and over again about companies encountering "quality fade" in China. They might get a 500GB flash drive the first few months... but then the drive becomes an 8gb drive with a controller lying about the rest and overwriting data soon after that. Or the titanium alloy needed for strength is replaced by some pot metal lying around the factory.
If I want it made well, the US, Europe, Germany, Japan, and Russia can make stuff to be proud of. Cheap? China is your go to party.
Re: (Score:2)
Nowadays it seems pretty much all non-trivial chips include an ARM core with enough storage and ram to run circles around any desktop computer found in the 80s or even early 90s.
In fact, ARM at 4MHz (reading ROM) or 8MHz (after copying ROM into RAM) was running circles around 80s desktops in 1987 - I used to emulate an IBM PC on an Acorn Archimedes [computinghistory.org.uk] at that time.
Re: (Score:2)
However the modern low power ARMs are Cortex-M series, which are very different from the more power hungry ARM from the 80s. Even the commonly used ARM7 is very high power in comparison to what is needed for an unattended battery powered sensor monitor.
Crazy Eddie's East (Score:2)
Now they throw the backdoors in for free!
Re: (Score:2)
A similar situation exists with WiFi - the ESP8266 [iot-playground.com]. These things can be bought for under $2 and include a 32-bit micro-controller with WiFi. Amazing.
Re: (Score:2)
The companies are jumping on the bandwagon. A couple years ago there were no good chips out there that met the needs of being battery powered for over a decade with low power wireless capabilities. Today there are a handful all in pre-release stages. There was a gap in the market and it's staring to be filled. The biggest snag for most is being low power, as in 1 microAmp or less sleep current, as that has not been a common feature request by customers until recently.
15K RAM is too small in my view if y
The obvious question. (Score:3, Funny)
Is systemD part of the OS?
Re: (Score:2)
If the size of the kernel becomes small enough, you could develop the binary by genetic algorithm and just evaluating it with tests + benchmarks :)
Re: (Score:1)
In 10K? I thought systemD had minimum system requirements of 8G
Re:The obvious question. (Score:4, Insightful)
Re: (Score:2)
Well that is your opinion which you of course are entitled to, however many in the embedded space are quite happy with systemd since it means that they can cut down on the number of old and unreliable services they used to rely on.
https://systemd.events/systemd... [systemd.events]
http://bec-systems.com/site/10... [bec-systems.com]
Does it really matter? (Score:2)
Does 10K vs. 200K vs. 5MB actually matter? RAM is so cheap, so small, and so ubiquitous today.
I appreciate that it doesn't need gigabytes of memory... but what does saving 190K of memory really get that wasn't possible before?
use non volatile memory (Score:2)
If that's the goal then why not just use non-volatile memory. Would that not save the same power without scrimping on quantity of memory?
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
At times we were considering using parts of Flash for temporary storage but that sucks up power too. 20-40mA while writing.
The scrimping is done for several reasons. It's all internal memory on the same chip as the CPU and that in itself limits the amount due to physical space of cramming everything in. External memory has it's own extra power needs with no capability of being put into low power standby, plus requires a lot more pins for interconnectivity, and adds to cost which is very important if you
Re:Does it really matter? (Score:4, Insightful)
IoT devices will typically only have a few kB of RAM, so yes, it does matter. As a data point, I'm working with an ARM Cortex-M0 processor at the moment and it has a grand total of 128kB program memory (flash, so you can't use it for read/write data storage) and 16kB RAM. This is an entirely typical processor for this sort of IoT application.
Re: (Score:2)
Why use an M0? Does it really save power? For many IoT applications it's a race-to-quiescent that matters. That is you wake up, do something simple and shut down. The longer it takes you to do something the longer one uses power. SO in the end it's how much energy the task takes not how fast you do it. And up to a point, faster systems often have better not worse energy performance per task. This is in part because ancilliary systems have to be power up too while the CPU is powered up and partly beca
Re: (Score:2)
Fast is sort of relative. The amount of state transitions in the chip determines most of the power, so running at a slower clock rate does not necessarily reduce power usage. All of the ultra low power chips I've seen have ways to turn off the clock to various modules or even power down entirely which solves a lot of that problem.
Cortex-M0 also has less die space overall compared to a Cortex-M3/M4, fewer transistors, smaller pipeline, and so on. So what it loses slightly in speed it makes up for in power
Re: (Score:2)
1MiB is huge in this subject (we are talking here in some KiBs)
Re: (Score:2)
True in terms of IoT where the system needs to run with minimal power, possibly on non-rechargeable batteries for a decade or two. But in the broader embedded world where devices are constantly the 1MB-4MB size range is pretty common. Though if you drop down to things like AVR or PIC then you're looking at less than 1K of RAM.
Re:Does it really matter? (Score:4, Informative)
Take a look at the specs on most microcontroller boards out there and you'll feel very differently about saving 190KB. In some cases that's more memory than the board comes with.
* TI Tiva Launchpad - 32 KB - http://www.ti.com/tool/ek-tm4c123gxl
* ST Discovery Board - 190 KB - http://www.st.com/web/catalog/tools/FM116/SC959/SS1532/PF252419
* Olimex STM32E407 - 192 KB - https://www.olimex.com/Products/ARM/ST/STM32-E407/open-source-hardware
* Arduino 101 - 24 KB - https://www.arduino.cc/en/Main/ArduinoBoard101
* Intel Galileo - 512 KB - http://www.intel.com/newsroom/kits/quark/galileo/pdfs/Intel_Galileo_Datasheet.pdf
I could go on, but you get the point.
Re: (Score:2)
Demo boards oftne use the higher end of the product line that has more RAM because it's more convenient for experimenting or evaluating. In actual production it's common to use the lower end of the line to save money or power.
Re: (Score:2)
640KB ought to be enough for anyone - apocryphal Bill Gates circa 1981. :)
Yes (Score:2)
Ever wonder why your current computer seems to run as fast as one from a decade ago? Shouldn't everything load much faster and be more responsive?
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
RAM takes power! When the device is in very deep sleep mode, any RAM that is retained has to be periodically refreshed and more RAM means more current to do that.
And this is just the OS remember, not a full system. The less that the OS uses the more the system and application can use. It is common for Cortex-M series processors to provide 16 or 32K RAM, where even taking up 10K for the OS use is sucking up a large amount.
Re: (Score:2)
The RAM doesn't have to be refreshed if it is SRAM, although it still needs power.
Re: (Score:2)
Sorry, it is "recharge" so I mispoke. And it is low leakage SRAM. The periodic recharge is not necessarily only for the SRAM but for all the components that are not completely powered down during a low power sleep mode (that is, voltage has not been cut). At least on the systems I've used, low power may be handled different on others.
Technical detail (Score:3, Interesting)
How fast does it power up? This is one area of computers that makes absolutely no sense. If something can be stored, why can't it be stored in a ready-to-run state? Why aren't drivers on ROM inside each device? My radio comes on instantly, so should the damn computer. We're back to the days of having to wait for the TV to warm up, shouldn't be that way.
Re: (Score:1)
The moment you set something in stone is the moment you realise you fucked something up.
Who said the chip can't be replaceable?
Re: (Score:1)
Even if wasn't a chip, one could always use a MicroSD card to stash firmware. An internal MicroSD reader is small and inexpensive, and can easily be made to be read-only. This might be a decent compromise between resistance from network/external attacks and the need to update. Of course, physical access can be an issue, but there is always requiring signed firmware and executables to deter that. If physical access isn't an issue, this might be an ideal way to upgrade firmware on devices that are somewha
Re: (Score:2)
The customer that you just shipped ten thousand of your products to, who just spend several months installing them all over the place. That phone call where you say "oops" can get pretty heated.
Re: (Score:2)
Re: (Score:1)
Doesn't matter. All it should need is standard input. The ROM does the rest.
Re: (Score:2)
Re: (Score:1)
The ROM is already matched to the CPU on the same board, or even on the same chip. Isn't that kinda what an SoC is? I'm not sure what you mean. Text/graphics to a printer or video to a monitor are the same regardless of the CPU. I only question the necessity of caching to RAM on every power up, it should just be nonvolatile with everything already there, instantly ready to run. ROM can store a fresh image if something breaks. Something is horribly wrong when my router, printer, computer, phone takes two min
Re: (Score:2)
Why do you comment on things you have no idea about?
What part of "Slashdot" don't you understand, boy?
Re: (Score:3)
What is the difference between 10kB and 1MB in price or volume?
The advantage of a minimal OS is it has less things in it therefore less things will break, so is more reliable but I don't see Linux as losing the reliability argument.
In my experience, it mostly comes down to what the difference is between on-chip memory and off-chip memory. I assume that nearly all basic IoT entities like your washing machine would use on-chip memory for reasons of both cost and size. And on-chip memory should be minimized to optimize cost and power. So, less is more, regardless of any sort of "less things will break" consideration. If I were a washing machine manufacturer, I wouldn't want to spend more than 50 cents on the whole thing.
Re: (Score:2)
... execution time?
Disk storage and/or RAM footprint is not the only thing that results when it comes to code size differences.
When you want to do something realtime, big code is your enemy.
All one address space :( (Score:1)
In this day and age of net threats, who on earth is going to expose something to the internet, whereby all processes including the kernel run in a single memory address space (i.e. no concept of rings or protected mode)? :(
It will be a potential security nightmare. I was quite impressed and excited until I saw that news about the memory map.
We used an older version of vxWorks in my previous job that ran that same way, and just chasing bugs down (that crashed the system hard or even overwrote the onboard fla
Re: (Score:1)
The current generation of ARM Cortex microcontrollers don't, as far as I know, include hardware a MMU. I agree that it'd be pretty awesome apply this sort of protection to embedded systems, but it seems that the hardware isn't ready for that yet, unless you achieve it in software using techniques such as static analysis, JIT, etc.
Just a multi-threading library (Score:4, Interesting)
This is really an 'OS' in the loosest sense. The processors it targets (mostly ARM Cortex M) do not have a memory manager, so all these things really do is provide a basic multi-threading system, which is actually pretty trivial to implement on a micro-controller if you can be bothered. I mean, a decent embedded unicode handling library or buddy allocator is a more complicated endeavour, but not as media worthy as saying your wrote a whole 'OS'.
For what it's worth, FreeRTOS has been providing a lightweight and portable kernel that basically does what this does for about a decade now. Like windriver, they also try to leverage the exposure they get from the free code to sell their commercial version.
The interesting thing though is that embedded processing power is really jumping along at an incredible pace now. To be honest, I think in another five year anything that requires connectivity or has a screen will just be using a Linux derivative (or QNX). Memory and processors are becoming ridiculously cheap, and in many cases the few dollars saved by using a resourced constrained chip are not worth the effort of having to port a graphics library, driver etc etc.
Re: (Score:3)
RTEMS.org already exists as a single process, multi-threaded RTOS with robust POSIX API support. Includes SMP, networking, filesystems, USB, high test coverage, etc. No dual licensing for commercial use. Used heavil by NASA, ESA, high energy physics community, and many industrial users.
Re: (Score:2)
Yup, FreeRTOS is a bit odd. It's also very large, which it makes up for with a bag full of #ifdefs to trim it back down. It's very difficult to read and follows one particular programmers personal coding style.
I think a lot of the portable RTOSes out there have a fundamental drawback that by trying to cater to so many chips portably that they end up being lower performance. Ie, in uC/OS they put a critical section around reading and returning a 32-bit value just because this isn't atomic on an 8 and 16-b
Re: (Score:2)
What is the difference between this, FreeRTOS, ChibiOS or any of the other open source RTOSes already on the market?
The biggest news here is that Wind River is opensourcing their RTOS.
Re: (Score:2)
Alright, let me try: Zephyr is no good because the ATmega328 Arduino is only 8-bit and only has 2KiB RAM.
What do I win?
Re: (Score:2)
Run NilRTOS [chibios.org]
Dont need an OS. (Score:3)
Running the application on the bare metal is just fine.
Sorry but very rarely do you need a full blown OS. my weather station does not need an OS, it just needs it's program to run and loop over and over and over again running directly on the hardware.
I can see things like Amazon Echo using an OS, but then it's processing power and capacity can easily run a full blown linux. no need for anything that is super tiny.
Re: (Score:3)
If it's cheaper than an 8 bit MCU? yes. I don't care if it has a 256bit MCU... it will never need an OS.
Re: (Score:1)
Well good to know, that when you don't need one, no one else needs either. Nice, one less thing to worry about.
Re: (Score:3)
I already have access to network devices, it's coded in the software, it has a wifi chip inside (ESP8266) and does perfectly without an OS or any drivers.
Not needed.
Re: (Score:3)
Re: (Score:2)
Yep, and a good programmer knows how to write a program that is multi threaded... still don't need an OS.
Apache 2.0 licensed (Score:2)
Feels like the license should have been in the summary. Whatever your opinions on copyleft in other areas, a GPLv3'd RTOS would be at a huge disadvantage adoption-wise. It'll be good to have an Apache licensed RTOS with some serious momentum behind it.
OS-9 (Score:1)
Microware needs to start porting OS-9 to these sorts of things.
I, for one, welcome the new IoT overlords! (Score:1)
Linux will be happy (Score:2)
Re: (Score:2)
Minimal error checking? (Score:2)
So, unified kernel/userspace. minimal error checking. Some error checkign available, but optional.
The Internet of never updated, easily pwn3d things.
license prevents drone use (Score:3)
look in the 'about' page, in the 'export compliance' section: "(c) for use in connection with the design, development or production of nuclear, chemical or biological weapons, or rocket systems, space launch vehicles, or sounding rockets, or unmanned air vehicle systems."
More reason to develop software ex-US, so crazy export compliance doesn't hobble you.
better alternatives (Score:4, Informative)
"The Zephyr Project's RTOS" is limited to x86 and ARMv7. x86 isn't for anything small and just using ARMv7 doesn't leave you with many options for chips.
Contiki [wikipedia.org] and ChibiOS/RT [wikipedia.org] have already been ported to run on damn near anything and yes, only still only require 10KB of RAM.
Firefox OS? (Score:2)
small footprint, modular, scalable
Those three things are mutually exclusive unless you change the definition so that it applies to basically anything on the planet.
Modular is never small, scalable generally isn't either.