Hacker News Reader

Comment by mariusz79 | Hacker News Reader

Latency is really not that big of a deal, as there are already protocols that deal with it quite successfully. And remember that slow Internet is better than no Internet.

View on HN

Latency is a big deal for some things. Although it's from 1996, the principle given here are still applicable:

    http://rescomp.stanford.edu/~cheshire/rants/Latency.html[0]

Discussed here:

    https://news.ycombinator.com/item?id=7826768[1]

And while it's true that slow internet is better than no internet, people still don't grasp that latency is a big deal - it really is.

I see that in this case the latency issue is being dealt with by using LEO satellites and doing clever things with hand-overs, and that's good. The question is whether it's enough, although given that it's Musk at the helm, it probably is.

Edited to add real links:

[0] http://rescomp.stanford.edu/~cheshire/rants/Latency.html

[1] https://news.ycombinator.com/item?id=7826768

of course for some things latency is important, but not for all of them.

No not for all, but perhaps for more than you think. Even simple web browsing.

Are you aware of just how many resources a typical modern web page loads, and how many requests it makes? As an experiment you could try putting a half second delay on every request and see what happens to your browsing experience. It's horrible, which is why I always worry about latency. It looks like Musk is avoiding it, but it needs considering.

Part of the problem is that web designers are largely protected from the effects of slow networks and bad latency, they design as if these things don't matter. And usually they don't, but it's becoming more important because they don't realise.

I remember computing when the speed of light didn't matter. Sometimes I miss those days.

Edit: Changed "Someone's" to "Sometimes" - auto-corrupt struck again.

That's an artefact of us not having to deal with that much latency, though, and is being fixed with the ability for the server to pre-emptively push resources you will need.

(as a side note, my first "web browsing" experience had a typical round-trip measured in hours; I requested my pages with CERN's email-to-web gateway, but I got my e-mail via UUCP exchange to a local BBS which again did their UUCP exchange to an internet connected host every 4-6 hours, so if I timed my exchanges right, I might upload my request right before an exchange, and get my page back right after the second exchange 4-6 hours later - of course it sucked, but it was still massively better than no web access, and at the time I had never experienced "live" internet access)

I think that people, especially web developers, tend to forget that the Web is not the Internet.

It sounds like you think I'm a web developer who thinks the web is everything. I'm not, and I don't. I work on safety critical systems doing hard-real-time and soft-real-time image analysis, object tracking, data correlation, and tactical overlays for widely distributed systems consisting of a mix of servers, embedded, and customer facing systems. More than one customer has come to me and said - "This is easy - just send it via satellite!"

I have dealt with and feel I have some understanding of the issues here.

Earlier you said[0]:

    "Latency is really not that big of a deal,
     as there are already protocols that deal
     with it quite successfully."

I'd be interested to know of which protocols you speak, and your experience in working with them. I'd be very happy to benefit from your experiences in this regard, as there's always a chance that you know things I don't.

[0] https://news.ycombinator.com/item?id=9327656

Don't make blanket statements without looking at the actual numbers first. The satellites will be in LEO. An added 5ms (or say 50ms with slack for demodulation, processing, etc) RTT is totally insignificant for their target market.

Their target market is unlikely to stay still, and once people have some internet access they will start to ask for more. Worse, suppliers will take it for granted that they have internet access, require that they use it, remove other methods of accessing information, and still ignore the question of latency.

See also my comment[0] about trade-offs between LEO and MEO, number of satellites versus number of hops, and other issues. I am aware of (some of) the issues, and it concerns me that many people seem to be less so and just assume that all will be well.

[0] https://news.ycombinator.com/item?id=9328145

Its only 127ms to geosynchronous orbit. So 256ms round trip. Not a half second. Not much different from going from USA to India.

The "half second" was as an example that's not too far off the right value when you add problems like congestion, retries, and similar, it wasn't intended to be exact. Even so, it's only off by a factor of two.

See also my comment[0] about trade-offs between LEO and MEO, number of satellites versus number of hops, and other issues.

[0] https://news.ycombinator.com/item?id=9328145

Its also right in line with existing web service delays i.e. USA to India. So nothing we haven't been dealing with for a decade now.

I get it; satellites have their challenges. But the delay is probably a small part of the issue.

Well, I don't know what latency the handoffs will cause, but 750 miles straight up is 4ms each way. Considering they'll often be at an angle, you might get roundtrip latencies of 10 to 15 ms. That's on top of anything else, but it's pretty good for most purposes.

Reply on news.ycombinator.com