Comment by kbenson | Hacker News Reader

Comment by kbenson | Hacker News Reader

Parent Full thread

> Heck people don't even seem to care much about cell phones and keeping them up to date (iPhone and Nexus aside).

I think that's inherently a feature of them being too closely linked with the companies that provide them. Phones are the extreme case of what IoT devices could become. Once the phone OS is no longer supported by provider, keeping it is both a liability, and increasingly problematic as things stop working. Compare to old Nokia bumb brick phones. People still use those.

While the close control over phones has yielded some advantages (a boost to their markets and security, for one), a phone is generally much more complex than the average IoT device, and is closer to a general purpose device.

> As for the last paragraph, perhaps certification instead of standards?

Certification doesn't prevent a service from going away a year later. We need a way to hook the device up to another service. If it's a standard, there's nothing to prevent an open source product that you can install on your own server, local or remote.

> While there could be conventions for REST endpoints, I'm sceptical they would be of any use. The devices by their nature are special purpose (rather than general) so there would be a lot more domain knowledge needed to make sense of them.

I would be interested in exploring some specific examples. What, besides aggregation of audio, video, or stats, and configuration options is there that would be hard to support? Also, I'm not saying we can't or shouldn't have specialized services where it makes sense, just that we should have open standards that companies can work to implement if they choose.

> Additionally making them robust against general clients versus only in-house ones is a lot more work.

They always need to be robust against general clients. Allowing companies to think they can get away with ignoring this will be very problematic in the future. Having a well defined standard and likely off-the-shelf software to help implement it will probably do quite a bit towards making things more secure, not less. See home routers and the use of openWRT, Tomato, etc. I think this would more than likely raise the lowest common denominator when it comes to security of IoT devices.

Note that certification could also include verifying there is a plan in place for whatever period of support is reasonable. (We could dig deeper on this.)

Configuration options are hard to support. Say for example a "power level" setting is exposed (how much power is consumed to provide the device functionality). But it is also the case that more power shortens the lifespan of the device. A general client can end up setting that to 100% but won't know about the consequent lifespan issues, or too low a level results in half the functionality being switched off. By robustness I don't mean malformed inputs etc, but rather what is done with valid values. A general client will be oblivious, but an in house could present only part of the power range (the rest behind an "advanced" section). Settings will also interact with each other in device specific ways (eg power would affect maximum volume for an audio device). Pretty soon you'll need a complicated highly featured description language for the settings, interactions, cautions, effects etc. Also would you want to take support calls if external tools have been modifying the configuration, given that a single support call costs the same or more than the profit on the device?

That all said, I see nothing wrong with someone publishing and maintaining a set of best practises. IMHO it is the route most likely to get traction and improvement in devices.

I think any certification that requires a plan be in place for future support is going to be either too weak (if the company goes under, support is gone) or too burdensome (requiring some trust be in place for future support should the company go under is very inefficient). Just allow more control by the user. This lowers barriers to entry for the market, and allows for interesting and unique future uses.

As for non-standard methods of configuration and support, I see no problem with requiring your own software if you want it to be supported, and requiring it be set to default levels (or a verification that no values are outside norms) before a support incident is allowed. You want to permanently log whether any settings have been set that void your warranty? Go ahead. This is a solved problem. Try to get Apple to fix your phone if the little water sensor inside has been tripped.

I see no problem certifying hardware, I'm just not sure how something like UL for software works in practice.

We likely have different ideas of what constitutes future support. You seem to want the A+ package. I think it is sufficient that consumers get a pro-rated amount of their money back, depending on time since purchase. That can be done similar to insurance or escrow. (Although that is still a form a trust for future support.) If you buy something for $100 and 24 months later they pull service, should you get back $10 or $100? Only the smaller end seems viable to me.

> Just allow more control by the user

Given infinite people, time and money that is easy. I can't see any way of it being practical though. It is considerably more engineering effort, considerably more support effort (all that stuff has to be documented, tested and revised), will increase time to market etc. The only winners are those folks still using the device when service is pulled!

I'm sceptical that generic third party configuration could be reasonably done well. Is there any example of something similar? SNMP was mentioned as a failure by another commenter.

Certifying the software could be done like some other certification is done:

* Publish the list of best practises (eg authentication, update mechanisms, how they participate in the network, long term service)

* Have auditors (yeah they would need to be certified too) that will come in and examine a product/code base/service and give their blessing for relevant best practises. I'd imagine some open source libraries would end up certified too (eg an updater, or authentication) so a quicker path to product certification would be use already certified components.

* Award the certification based on auditor report

Yes it could be gamed, there would be wiggle room, bad actors etc. But it would be a lot better than we have now, and a feedback loop would help improve.

BTW I am working on a product somewhat in the space. I'd love to make it completely open, available for ever, plugin your own servers etc. But it isn't remotely viable for people, time and money reasons.

I am not by any means an engineer so you can most definitely take what I say with a grain of salt. But my understanding is that all that really needs to happen for IoT devices to be usable by a third party is for the manufacturer to simply publish the API and provide a mechanism for the end user to control what interfaces with the API. It doesn't actually require you to standardise the API as you have never guaranteed compatibility with any third party it simply requires documentation, which would have to already exist internally for development purposes, and on some level some loss of control. But surely, this way you then get to maintain control and development for as long as you want to support it and for those people that then want to support it themselves after you have abandoned it can as API is a known entity. I am probably fundamentally misunderstanding an aspect of the development of this somewhere but this does not seem that burdensome to me I don't even see the necessity of a standard that needs maintaining here just documentation of the API and a mechanism to change where the data is sent. Hell if you want to maintain control for the life of the device fine, but what is so hard about releasing that information when you no longer want anything to do with it, what possible commercial value does that information hold then?

> ... for the manufacturer to simply publish the API ...

That is more work than not publishing the API. How much more work then? I'd pick stats from The Mythical Man Month where they claimed that doing something for external consumption is nine times the effort than internal only. Admittedly it does predate AWS, containers etc, but then again those aren't that useful for iot.

Ok makes some sense but why not just provide the API and a mechanism to change where data is sent after you have decided not to support the device anymore? Surely that is not to burdensome and will give someone out there the option to continue using the device (and probably save some goodwill over the dropping of support)?

> You seem to want the A+ package.

No, I'm fully on board with zero support, given that's communicated up front. Let the market sort out whether people think that's a good idea. I care less about certifying devices to some standard level of quality, mainly because we're so far away from knowing what that is, than I am in opening up the ecosystem to fight planned (and negligent) obsolescence.

> I'm sceptical that generic third party configuration could be reasonably done well. ... SNMP was mentioned as a failure by another commenter.

SNMP is still used quite heavily in network devices. It's painful because it's complicated, but it's still one of the standards you can rely on any enterprise network device to support (as I understand it, I've dealt with that less the last five years or so), at least for the reading of information. In that respect, it's a success story. That said, I imagine the average network device in that category might be an order of magnitude or harder to deal with configuring (physical connector options, link level options, protocol options, route control and propagation, access lists, etc).

> Is there any example of something similar?

IP, TCP, UDP, POP3, IMAP, HTTP. The core protocols started as free standards, but for mail, compare to exchange. Exchange persists still due to its advanced capabilities and tight integration with other Microsoft products, but it's not what ended up being used for integration in the current generation of mail programs. They all support IMAP and/or POP3, because why write your own tools to migrate mail or provide third party access when a suitable protocol and suite of tools exist.

> BTW I am working on a product somewhat in the space. I'd love to make it completely open, available for ever, plugin your own servers etc. But it isn't remotely viable for people, time and money reasons.

Yeah, and I'm not arguing that you should do this. But if a standard existed, and there were free or cheap libraries that did most the heavy lifting for you, it's possibly that would make it the more viable choice for those same reasons. E.g. A light weight HTTP server with route engine to hook into your own libraries,or even to just write to a config file and HUP your daemon.

Reply on news.ycombinator.com