Uh, how often are you using the Internet to connect to a computer in your home town? Maybe 5% of the time?
I’ve never used Starlink, but with a basic understanding of geography and optics, I’m going to bet that in most scenarios the latency difference between Starlink and fiber is negligible, sometimes even being faster on Starlink, depending on the situation.
That said, I’m not suggesting Starlink is a realistic replacement for fiber, just that latency isn’t the big issue. (It has other serious issues)
Ok, so actual question, How useful are CDN endpoints these days with https everywhere? Because most encrypted content is unique to a single web user, caching isn’t super useful. Also you can’t cache live content like video calls or online games. I’d imagine the percentage of cacheable content is actually fairly low these days. But like I said, I don’t actually know the answer to this, i’d be curious to hear your take.
Edit: it’s weird to get down votes for a question.
HTTPS / TLS has little to do with it. Don’t think of the endpoint as a cache between you and the origin. The DNS name given to the endpoint is the origin from your browser’s perspective. How content gets cached on the backend is irrelevant to the browser.
Live video that someone else in your area is also watching is cacheable. Images to load a page, very cacheable. The personal stuff is mostly HTML specific to you but that’s quite small.
Browsers partition the cache by “origin” now though, so while it can cache HTTPS content, it can’t effectively cache shared content (It’ll store multiple independent copies).
So Youtube still works fine, but Google Fonts is pointless now.
Edit: Oh yeah, and any form of shared JavaScript/CSS/etc. CDN is now also useless and should be avoided, but that’s always been the case.
Yeah, this is the point I was getting at, encrypted content tends to be personalized to individual users, not always I guess. But yeah, I’m not sure how much is left.
It was an issue for a long time that browsers just ignored the caching headers on content delivered over HTTPS, a baked in assumption that they must be private individual content. That’s not the case now, so sites have to specifically mark those pages as uncachable (I think Steam got hit by something like this not that long ago, a proxy was serving up other peoples user pages it had cached).
But for something like Google Fonts, the whole point of it was that a site could embed a large font family, and then every other visited site that also used it would simply share the first cached copy. Saving the bandwidth and amortizing the initial cost over the shared domains. Except now that no longer holds, instead of dividing the resources by the amount of sites using it, it’s multiplying it. So while a CDN might put the contents physical closer to the users, it doesn’t actually save any bandwidth (and depending on how it’s configured, it can actually slow page loads down)
I live near DE-CIX and have fiber. So a decent chunk of web services I use is available with a latency of under 5ms. And everything else hosted in a European datacenter with under 20ms.
So almost all of my internet traffic has a lower latency than starlink has under ideal conditions
The point is, unless you’re playing some hyper competitive game where a 30ms difference in reaction time is noticeable (this is less than 1 frame in a fighting game, for example) Starlink works perfectly well. Lower numbers are better, but for games you only need to compare that number to human reaction times (150-200ms) to see that both are small values less than the reaction time of any person.
Previous satellite Internet using satellites in geosynchronous orbit had 1500ms latency, for comparison.
where a 30ms difference in reaction time is noticeable (this is less than 1 frame in a fighting game, for example)
You have some pretty bad understanding of how netcode works if you think a 30ms ping in an online multi-player game means your game or input is delayed by 30ms. It’s a lot more complicated than that, and especially in games with bad netcode you will absolutely notice a difference between 10ms or 30ms ping
Previous satellite Internet using satellites in geosynchronous orbit had 1500ms latency, for comparison.
Yes, and are far more stable, not hyped, and are already at pretty much peak congestion. Starlink will get progressively worse, the more people use it. Right now, it’s over provisioned.
The point is, unless you’re playing some hyper competitive game where a 30ms difference in reaction time is noticeable (
Yes, and are far more stable, not hyped, and are already at pretty much peak congestion. Starlink will get progressively worse, the more people use it. Right now, it’s over provisioned.
They were not more stable. Any occlusion, including thick clouds, would degrade the signal to being unusable. I used Hughsnet for years, then swapped to cellular (100ms+ latency) and finally to Starlink. Starlink is a pretty solid 100Mb/s, with low jitter, packet loss and latency.
Ever try a voice call with 30ms of latency?
Yeah, I use voice chat every day, it’s not noticeable.
Yes, your understanding is fundamentally flawed. Starlink adds a fixed latency on top, if you ping to a server was 2ms with fiber and 52ms with starlink, then your ping to a server that would be 100ms with fiber would be 150ms with starlink
Of course they don’t mean in every case. Yeah, if you have to go halfway around the world from two addresses that are very far away from hubs, Starlink might be better. 99.99999% of the time this isn’t happening though and fiber will be better. There are situations for some people where it’s worth it. Fiber is better for the average case though, and it’s where money should be invested.
You’re probably really far away from the VR Chat server. Try pinging Google or Cloudflare, which will tell you ping to the nearest datacenter (a rough estimate of ping caused by your local ISP).
Based on their numbers, you could probably expect 50-100ms to Google, and then add an extra 90ms to get from there to your VR Chat server.
My personal fiber connection gets under 2ms ping on Speedtest
That makes sense then. When people talk about their ISP ping, they’re usually talking about how long it takes to get out of the ISP’s network. So that 5ms Cloudflare ping is likely pretty close to what people would consider your internet’s ping.
Speedtest.net is a really common tool for measuring this, since it will automatically check where the closest server is. For your connection, any ping above 5ms you can probably assume is based on your physical distance to the server, or latency on the server’s end. I’m guessing Google doesn’t have a server quite as close to you as Cloudflare
Thanks for the details! This makes sense now. I started asking questions because it seemed wild that the only ping I pay attention to, the one shown in a game I play, would be up to 4.5 seconds on starlink. I guess it would be ~250ms at the top of the range they quoted.
Starlink has much better latency than most satellites, but still 10 to 50 times as much as fiber.
ha yeah… not having to make a 340 mile round trip instead of the hundreds of feet to the nearest router will do that
Uh, how often are you using the Internet to connect to a computer in your home town? Maybe 5% of the time?
I’ve never used Starlink, but with a basic understanding of geography and optics, I’m going to bet that in most scenarios the latency difference between Starlink and fiber is negligible, sometimes even being faster on Starlink, depending on the situation.
That said, I’m not suggesting Starlink is a realistic replacement for fiber, just that latency isn’t the big issue. (It has other serious issues)
Living right near a massive CX that services the US-Canada border… most times.
Much more frequently than you think with CDN endpoints.
Ok, so actual question, How useful are CDN endpoints these days with https everywhere? Because most encrypted content is unique to a single web user, caching isn’t super useful. Also you can’t cache live content like video calls or online games. I’d imagine the percentage of cacheable content is actually fairly low these days. But like I said, I don’t actually know the answer to this, i’d be curious to hear your take.
Edit: it’s weird to get down votes for a question.
HTTPS / TLS has little to do with it. Don’t think of the endpoint as a cache between you and the origin. The DNS name given to the endpoint is the origin from your browser’s perspective. How content gets cached on the backend is irrelevant to the browser. Live video that someone else in your area is also watching is cacheable. Images to load a page, very cacheable. The personal stuff is mostly HTML specific to you but that’s quite small.
HTTPS can in fact be cached, and most modern browsers will do so unless given a header or something to tell it not to.
Source: Devtools network tab + https://developer.mozilla.org/en-US/docs/Web/HTTP/Guides/Caching
Browsers partition the cache by “origin” now though, so while it can cache HTTPS content, it can’t effectively cache shared content (It’ll store multiple independent copies).
So Youtube still works fine, but Google Fonts is pointless now.
Edit: Oh yeah, and any form of shared JavaScript/CSS/etc. CDN is now also useless and should be avoided, but that’s always been the case.
Yeah, this is the point I was getting at, encrypted content tends to be personalized to individual users, not always I guess. But yeah, I’m not sure how much is left.
It was an issue for a long time that browsers just ignored the caching headers on content delivered over HTTPS, a baked in assumption that they must be private individual content. That’s not the case now, so sites have to specifically mark those pages as uncachable (I think Steam got hit by something like this not that long ago, a proxy was serving up other peoples user pages it had cached).
But for something like Google Fonts, the whole point of it was that a site could embed a large font family, and then every other visited site that also used it would simply share the first cached copy. Saving the bandwidth and amortizing the initial cost over the shared domains. Except now that no longer holds, instead of dividing the resources by the amount of sites using it, it’s multiplying it. So while a CDN might put the contents physical closer to the users, it doesn’t actually save any bandwidth (and depending on how it’s configured, it can actually slow page loads down)
Netflix and Amazon Outpost makes it quite useful.
I live near DE-CIX and have fiber. So a decent chunk of web services I use is available with a latency of under 5ms. And everything else hosted in a European datacenter with under 20ms.
So almost all of my internet traffic has a lower latency than starlink has under ideal conditions
Just for reference, I get about 45-50 ping playing Marvel Rivals on Starlink.
On fiber, while I don’t play that game, I’ve never seen a ping longer than 10-13msecs.
The point is, unless you’re playing some hyper competitive game where a 30ms difference in reaction time is noticeable (this is less than 1 frame in a fighting game, for example) Starlink works perfectly well. Lower numbers are better, but for games you only need to compare that number to human reaction times (150-200ms) to see that both are small values less than the reaction time of any person.
Previous satellite Internet using satellites in geosynchronous orbit had 1500ms latency, for comparison.
You have some pretty bad understanding of how netcode works if you think a 30ms ping in an online multi-player game means your game or input is delayed by 30ms. It’s a lot more complicated than that, and especially in games with bad netcode you will absolutely notice a difference between 10ms or 30ms ping
Yes, and are far more stable, not hyped, and are already at pretty much peak congestion. Starlink will get progressively worse, the more people use it. Right now, it’s over provisioned.
Ever try a voice call with 30ms of latency?
They were not more stable. Any occlusion, including thick clouds, would degrade the signal to being unusable. I used Hughsnet for years, then swapped to cellular (100ms+ latency) and finally to Starlink. Starlink is a pretty solid 100Mb/s, with low jitter, packet loss and latency.
Yeah, I use voice chat every day, it’s not noticeable.
That’s basically perfect, with regards to online gaming.
I got better ping playing Quake multiplayer in 1996
Online and not LAN? I have doubts.
That used dedicated servers, right?
So if my ping is currently 90ms on fiber, it’ll become 900ms - 4.5s on starlink?
Probably no. Your ping is abnormally high for fiber, I’d expect a sub 10ms ping for you.
That makes a lot of assumptions about what I am pinging, and the networking context.
In my case I was quoting my average ping in VRChat.
How can you quote 10-50 times higher and then tell me no when I calculate what that means for me?
Is it because latency does not scale in that way?
Yes, your understanding is fundamentally flawed. Starlink adds a fixed latency on top, if you ping to a server was 2ms with fiber and 52ms with starlink, then your ping to a server that would be 100ms with fiber would be 150ms with starlink
tracerouteliketraceroute cnn comctrl-cat the third line.Don’t try to ping UK.battle.net or your numbers will be skewed by everything in between.
About 5ms.
Based on the various replies, it sounds like the poster I was originally replying to does not mean pings in any context.
They just mean in this context. Along optimal routes. Right?
Of course they don’t mean in every case. Yeah, if you have to go halfway around the world from two addresses that are very far away from hubs, Starlink might be better. 99.99999% of the time this isn’t happening though and fiber will be better. There are situations for some people where it’s worth it. Fiber is better for the average case though, and it’s where money should be invested.
So then 10x makes 50ms; sounds about right
Of course. Still, an exception doesn’t disprove expected averages.
So you were only talking about when testing with ideal servers? Why is my example an exception? Are all games an exception?
Because we’re talking about the inherent latency of the connection, obviously.
How condescending. I’m obviously not wise to networking stuff. That’s why I was asking questions.
You’re probably really far away from the VR Chat server. Try pinging Google or Cloudflare, which will tell you ping to the nearest datacenter (a rough estimate of ping caused by your local ISP).
Based on their numbers, you could probably expect 50-100ms to Google, and then add an extra 90ms to get from there to your VR Chat server.
My personal fiber connection gets under 2ms ping on Speedtest
It depends on the instance (people can make them in 4 regions of the world) but 90ms is common for US west and east, for me.
That makes sense then. When people talk about their ISP ping, they’re usually talking about how long it takes to get out of the ISP’s network. So that 5ms Cloudflare ping is likely pretty close to what people would consider your internet’s ping.
Speedtest.net is a really common tool for measuring this, since it will automatically check where the closest server is. For your connection, any ping above 5ms you can probably assume is based on your physical distance to the server, or latency on the server’s end. I’m guessing Google doesn’t have a server quite as close to you as Cloudflare
Thanks for the details! This makes sense now. I started asking questions because it seemed wild that the only ping I pay attention to, the one shown in a game I play, would be up to 4.5 seconds on starlink. I guess it would be ~250ms at the top of the range they quoted.
My average latency on Starlink over the past year is 32 ms. It varies throughout the day from around 20 to 40 ms.
If you are getting 90ms on fiber, you are either pinging a server that’s a long ways away or something is very wrong.
If you look at the rest of the comments, you’ll see I was taking about my ping in a game. Not my shortest path to a nearby server.