Hacker News new | past | comments | ask | show | jobs | submit login

As to other reasons apart from the violation of privacy: Every network call adds additional latency and slows down interactions with the OS. Every data gathering feature adds additional complexity to the implementation, takes attention away from other implementation work that could be done instead, and increases the risk of adding further mistakes to the implementation. Personally, I would like OS and application vendors to work on improving security and correctness of their programs and reducing latency instead of adding data gathering features.

> But the line between a necessary network call and an optional one is often blurry.

What would be an example of a necessary network call that an ideal OS (i.e., one that cannot be easily compromised and does not require updates around the clock to correct programming mistakes) has to perform on its own?

If a company is interested in how users use their applications and desperately need our data for it, they may be interested in funding dedicated studies and appropriately compensating users that send their data, if it is so valuable for the company.






> What would be an example of a necessary network call that an ideal OS [...] has to perform on its own?

Syncing the clock with NTP?


> Syncing the clock with NTP?

So every app, instead of querrying the OS, shall make a network call, to get the time from an NTP server ?


I don't understand how you got that. The question was about network calls that an OS would perform on its own.

Have you developed large applications with/without anonymous usage data?

You need a good volume of data and you aren’t going to want to pay for it for one simple reason: you can get it for free and only a tiny group of users are going to be upset enough by this.

Not sure what the reference to “ideal OS” is about. I thought this was about windows in particular.

Necessary network calls would be related to updates, licensing etc. But the thing is: they would be going “home” to the exact same servers as telemetry AND they would easily contain the same payload.


> You need a good volume of data

it is called testing. _Testing_. But of course, testing sucks and it's expensive.


Testing?

You can’t say how your users use your software through testing. Not by surveys/panels/interviews either.

But yes: alternatives are also morr expensive (which means it’s expensive for the end user). Users pay one way or another.


No, sorry. Testing answers “does the feature work?”. Usage telemetry answers questions like “was the feature a good idea?” and “are enough users successfully using the feature to justify the cost of creating/maintaining it?”.

Those are not questions for which pre-release testing can provide answers.

I’m not weighing in on opt-in vs opt-out, or on anonymization. Just saying that testing doesn’t cover this niche.

(Separately, I think you’re largely wrong about testing as well: crash dump collection is about finding issues that pre-release testing wouldn’t find at any price. For things like OSes especially, the permutation space of hardware * software * user behavior is too large. While I’m sure a few companies use crash reporting as a crutch to support anemic QA programs, I do not think that many do.)


> Usage telemetry answers questions like “was the feature a good idea?”

You mean you implement something even if nobody asked for it ? Wow.

The project really has some spare budget.


That people ask for it doesn’t make it a good idea. Even if it’s a good idea and people asked for it that doesn’t mean people used it because they might not know about it. Is the feature prominent or intuitive enough? Should it be described in some newsletter or documentation?

You will never know without actually asking enough users (which is a large sample). And there is a simple way of ”asking” this.



> What would be an example of a necessary network call that an ideal OS

DHCP


Not strictly necessary though.

I grew up sans DHCP with static IP assignments per device .. and still practice that on modern home networks and production networks.

The only DHCP calls here are made by foreign devices wanting an assigned address, which gets them on a narrow range on a side net.


In that case sure maybe not. However, most systems aren't run by deep experts but by regular users which expect a device to be plugged into a network and then have the capability to use the internet without user interference. That more or less necessitates DHCP.

> However, most systems aren't run by deep experts ...

Luckily static IP addresses can be set up by the majority of teenagers that just want to play Doom, etc.

At least that was the case decades ago .. is this now "deep knowledge" that necessitates that OS's have to use DHCP with no other option ?

Perhaps we have different understandings of the words "necessary" and "sufficient", etc.


This and also it's pretty obvious that the main goal of both Microsoft and Google is NOT to make the OS better for its users.

So the claim that telemetry is used to improve products is simply a lie IMO.

The fact that telemetry is sent at all for no apparent reason and deliberately without clear consent is an ironic example of this. The fact that it's been happening more and more over the past decades as the OS'es evolved is another confirmation of it.


I generally agree, though for system settings specifically, I wonder what kind of ad targeting would you get out of that?

Still think it shouldn't be there by default - it reduces privacy and is a lame excuse not to do (paid) user studies.


> for system settings specifically, I wonder what kind of ad targeting would you get out of that?

You get sensitive data out of system settings, such as for instance health data: Does the user have a vision or hearing impairment, use assistive technologies etc.?


> (paid) user studies

Would it count as a paid user study if enabling telemetry for Windows knocked $10 off of the price of your computer?

I can’t decide if that’s a neat idea or dystopic. Which, historically, probably means it’s dystopic and that plenty of people are already doing it.

I think “traditional” paid user studies often suffer from the same sampling problems that make political polls and behavioral paid medical studies less useful (you’re not surveying the average voter; you’re surveying the average voter who likes to answer polls). But maybe the “$10 off” idea would capture a broad enough demographic as to be more useful.


>Every network call adds additional latency and slows down interactions with the OS.

...unless they're done asynchronously




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: