Best Platform / Module for production-ready commercial use-case?

Hi!

I am currently working on integrating a resin-powered device into an existing commercial hardware and have made some first steps on both Raspberry Pi 3 and the Intel Edison.

In my eyes the Raspberry Pi 3 worked more smooth in booting and serving the deployed app which was a basic Flask-powered Application with asynchronic network library and web-socket via Socket-IO.

Deploying the same application onto the Intel Edison worked without a problem but much slower and with much more delay. Also the web-terminal is a whole lot more slower when using it for the Intel Edison.

For my project however I would prefer using the Intel Edison over a Raspberry since this will be released as a commercial product and using the Raspberry Pi here as the resin platform feels a bit strange.

Also I do not need USB-Ports and physical Ethernet, just Wi-Fi and two UARTs which are used to communicate with the existing hardware.

So I was asking myself what other people, who are using resin.io - especially in commercial products - chose as their platform for resin.io!?

Is the Intel Edison a good choice for a production-ready platform? Are there any better options, like the Samsung Artik 5?

I have been thinking of using the upcoming Raspberry Pi 3 Compute Module but since there aren’t any facts available yet I strongly think that this won’t feature a wireless connectivity option.

So I really prefer using modules that come with on-board Wi-Fi so that this will minimize the effort which will be needed to build a interface board for that module.

Any help and feedback here is greatly appreciated!

Thanks,

Simon

PS.: I am still flashed by how awesome resin.io works! I really fell in love right after the first time I ran my just prepared Raspberry Pi and right after it had appeared in the resin.io cloud overview! So well done, folks! Keep that good work up!

I have been making a test and analyzing the Ping / Pong latency of a simple message thats been send out by the resin-device to the client (browser) over web socket using socket-io.

The Raspberry is always ahead of the Intel Edison. The subjectively felt slower reaction of the Intel can also be seen in these values.

hey, yeah, I’m not really surprised about this result, the Raspberry Pi should have more computing power onboard, even if considering ARM vs x86, the raw MHz and core count is in its favor. It’s quite difficult to rely on unreleased/planned boards indeed, and as you say, better to work what’s available now.

Comparison with existing hardware is a bit tough, because everyone’s requirements are somewhat different. For example, if you had to make a list, what would be your hard requirements for the boards you need to use (must have specs), and what would be nice to have? Such a list would help you sort through potential devices easier.

Some points of reference:

  • I know a few customers, who use Raspberry Pi (not even all of them the RPi3 but earlier) for sure (for example, I spy RPi in the NVBOTS video)
  • we had a drone demo not so long ago, using resin, and that uses the Dronesmiths Luci board, based on Edison (mostly for size, I think). Here’s a demo and writeup for context.
  • I know there are other boards coming out too, that are not yet supported on resin, but could be done as DIY device support possibly, or who knows. For example the Intel Joule board (Joule’s comparison with Edison)
  • ARTIK devices should be indeed a possible choice if the Edison/RPi3 are. If you are making a commercial product, the best is to get some boards for yourself and try them out. Same goes for any pottential board. All of them are pretty cheap compared to the technical dept a wrong board will give your company

By the way, is your test code available somewhere? Would be interested in trying it out, just for comparison :slight_smile:

Hey!

I’ve already tested the Intel Edison vs. Raspberry Pi 3 and although I really thought that the difference between those two were not that high - mainly because of the fact that the tested application is very small and lightweight - the Intel Edison is definitely slower than the RPi.

I’ve ordered an ARTIK 5 Dev-Board today and will make some tests and get back!

Here’s the source:

https://github.com/simmikolon/resin-flask-socketio-pingpong.git

Hi Simon,

I am wondering what was your final conclusion between Edison, RP and Artik? I am trying to understand the factors that would sway developers to one or the other.

Thanks,
Usman

1 Like

Well it totally depends on what you are trying to do. I am working with both the Edison and RPi now.

RPi has:

  • lots of processing power
  • Integrated WiFi (RPi3) and BTLE
  • Integrated wired Ethernet
  • Lot’s of community sources, article, etc.
  • HDMI, Display-Connector

So the RPi in my case is the choice when making stuff like Gateways, Industrial Controllers, etc. Everything that needs a display, etc.

Edison has:

  • A much much smaller form factor
  • Low power
  • No wired Ethernet - But it is possible to add wired ethernet using USB
  • Integrated WiFi, BTLE
  • And 4GB of eMMC Flash
  • Lot’s of IOs
  • Well subjective but you have to count it in: “Intel Inside!” - That still counts!

Edison is the choice when making stuff like wearables, wireless, battery-driven devices that come headless, for example.

Both are great devices and both really have performance. Very often I am just using 5% of the RPis resources.

My first impression of the edison being a slower devices was wrong. The benchmark was more a benchmark of the Resin.io tunneling / public-url feature - not about the device itself. I have been making my own tests and hit quite the same results when benchnmarking low HTTP Request Counts on both devices.

Simon

2 Likes

Thanks Simon! This is very helpful. So both the performance and choice are dependent on use case I suppose. And I assume cost is not a big difference here, and perhaps people working on scale are not too concerned with cost.

Usman

Intel has discontinued Edison
https://software.intel.com/en-us/iot/hardware/discontinued

And that is the result
https://communities.intel.com/thread/115536

I guess currently (2018) we should look for other platform.

Regards,
llap