"Unlimited" ISP plans

Allow me to give you guys a couple of hints:

  1. Peering: every ISP doesn’t have to be peered with every other ISP. What happens is ISPs are peered with IXs (Internet Exchanges). One of the biggest in Europe is AMS-IX (Amsterdam Internet Exchange). Every ISP worth its monthly subscription is peered there. So to get to a client (and vice versa), if our two ISPs aren’t directy peered, my connection is bounced through the IX. Is it faster than direct peering? Of course not. Is it simpler to get all the European ISPs interconnected? Yes. What usually happens is ISPs are peered with IXs in countries they see a lot of traffic in. Goes without saying that every ISP in any country should be peered with their own country’s IX.

  2. Symmetric vs Asymmetric: Ideally we would all live in a perfect world that had 10Gbps symmetric to every home. The world has to live on what is in the ground/sea though (you’ll get it) and the ISPs are not too keen on upgrading their infrastructure. If you are in the US for example, the ISPs are actively campaigning/lobbying to not upgrade. The download speed is easier to oversubscribe than the upload speed because not everyone will be paying attention to it all the time. If it downloads in 6 minutes instead of 5, nobody will notice (except that one guy that speedtests 24/7/365 and is always "but I’m paying for 100Mbps and yesterday at 21:00 I only got 89Mbps for 52 seconds, and usually drives a BMW). Upload on the other hand, different story. If you send a huge file, you expect it to go as fast as it can go.
    This brings us back to “what is in the ground/sea”. If the ISP is using the upload capacity to run its own services (colocation,hosting,vps,etc…) then he can’t easily oversubscribe upload. So the ISP has two choices: upgrade infrastructure, or limit uploads. Which do you think is easier/cheaper to do? Hire one of the 7 ships that specialize in laying undersea cables (@~$50,000/day), or say “here is your brand new 1000 down/100up connection”? I’m not talking about and replacing a 2km peering line. I’m talking about laying 2000km of undersea cable back to the big IX. Of course there will always be that one guy that replies “but why not replacing the transceivers on either end and be done with it”. Simply because that fiber wasn’t designed for these transceivers.

Another reason is that some of the technologies used or subscriber connections are asymmetric. GPON has ~2gbps of download, but only ~1gbps of upload per port. DOCSIS and ADSL also favor download at the expense of upload.
Also, most subscribers care only about download. How fast does this youtube video load, how fast I can download some file. If you are uploading a youtube video, it may not matter that much if it takes two hours instead of one.

I manage the networks of a couple of small ISPs. Those ISPs provide symmetric connections but the download traffic is way higher than upload traffic.
OTOH, single mode fiber can do 10gbps, probably more. It’s the matter of swapping the equipment on both ends of the fiber. Then again, the vast majority of residential users do not need it (yet), so why do it?

I know they are asymmetric, I just gave the “behind the scenes” reason as to why the technologies where actually designed that way ;-). I could go into “there are more carriers used for download, than for upload, so the total throughput ends up being more in the download direction than in the upload direction”, but that’s getting into the technical snoring details :slight_smile:

In a perfect world, yes, fiber is laid once and it’s done. It’s all transceivers from there. How about when you discover that the cable you ordered and laid 2000km of it, doesn’t 100% meet the specifications/conditions? We’ve had that happen, we ordered dropcables that were supposed to be heat resistant that ended up sagging and caught by trucks. Did the manufacturer “cheat”? No. Were the specifications off? No. The reason was that temperatures way overshot (+5C) because of the summer. That made the cable “pliable” enough to sag down and be caught by tall trucks.

Not to mention the glue that holds together ribbon fiber. It would get old and the fibers would fray from each other making the system more or less impossible to maintain or repair.

And one of the main reasons for that is that normal people care more about download than upload. If you offered a 10mbps symmetric line vs 30mbps download/2mbps upload, most people would choose the asymmetric option. Especially since with some of those technologies it’s not 1:1 tradeoff - you gain more download by sacrificing some upload. Even on an ISP that offers symmetric connections (because they just use Ethernet) people mostly download files and rarely upload them. So, it makes sense to optimize a technology for more download than upload.

I only have experience with small ISPs, so it’s different than a very large ISP.


Salt direct due to the iXs:

AR-IX Cabase

Mountain West Internet Exchange



Probably direct peering with valve e.x

Wrong. BTW salt used to have only one (yes one, as in 1) peering provider!
The whole traffic was sent over Cogent. Cogent had a Facebook peering issue so all SALT customers could not use FB or insta for a day :smile:

They have become better but are by far still the worst ISP. Even their core switches are grossly overbooked. 128 customers share one 25GBit uplink on the same switch in core.

Surely that is a bonus? :wink:


They are the only ISP you mentioned that is peering directly on a iX according to peeringdb. Is peeringdb wrong? :slight_smile:


Probably not.

But what if peering isn’t the fastest option?
What if other providers use cache servers from Steam in their core?
How fast ist the peering? How many customers share that line?

Well if you peer on 4 different iXs and the other is not, your core network 99% can handle more traffic since you are not paying transit fees for downloading games e.x unless they have shit design. Makes no sense, so i would bet my money on salt just looking at peeringdb.

Off-topic: I my area most apartments are wired for 10Gbps fiber. If people actually knew their neighbors these days all it would take is one homelabber in each building (or even block) to run a Steam Cache and everyone would benefit.

But… it’s rare to know the person who even lives across the hall from you these days… :frowning:


Then you would loose money :smiley:

Well they have 2x 100Gbit links 1x 20gbit and a 1x 10gbit on those IXs.

So a total of 240GBit. How many customers will share that pipe?
240 customers so everyone gets 1GBit?

Yes overcommit all ISP does that to private end customers. Only companies has dedicated fibers.

Sure. But some overcommit more. Some less. Some have more customers, some have bigger pipes. Some even have cache servers.

But I won’t drag it out any longer. The correct answer for Cyberpunk 2077 launch day was:

Last place with a measly 1MBit was B.

No that’s not the reason. Unless your ISP has racked a Netflix PoP cache or a Google one. Apple, nah. But what do I know?

Of course my ISP has racked multiple different PoP/core caches.

The data is coming not from SLC, but from the libuplink somewhere else. You cannot be close to every customer in the world. The satellite doesn’t proxy data, it’s an orchestrator.