[bb88]

Latency is different than bandwidth.

I can make a very low latency connection but only delivers at most 1 packet every 10 seconds. It responds in 5 milliseconds say, but it won't let you have another packet for 10 seconds.

Or I can have a high latency connection that delivers a huge amount of bandwidth. It might take 10 seconds to start the firehose, but once it starts, it's a firehose. This is ironically the problem that the internet has with "Bufferbloat", and high bandwidth radio connections in space.

A topological solution would be counting the number of hops. There are internet protocols that do this, it's just that I don't think that BT takes advantage of it.

[zzzcpan]

When bandwidth was a big concern bittorrent trackers prioritized peers geographically and ISPs were using retrackers to force peers within the same ISP to save bandwidth. It was very efficient, each torrent was downloaded pretty much just once for an entire ISP from the fastest sources. Good times.

But none of this is necessary, fast clients simply send more data and dominate download bandwidth. Doesn't matter why they are fast, whether they are simply very close or just on a path with more capacity. Everything is naturally efficient.

[bb88]

> Everything is naturally efficient.

Bittorrent is not an effective means of serving data from a 4G device. In fact, if anything the bandwidth on 4G arguably costs more than bandwidth from your Fiber ISP.

We use services like Youtube and instagram to push our data to others. And the efficiency from the phone perspective is the same as Multicast because the phone sends data exactly once over the upstream link.

You could say, "Well, duh, nobody should use bittorrent over 4G because of bandwidth issues." And that's exactly the point. If Bittorrent were so efficient, we'd be using it on 4G instead of sending it to youtube, etc.

You can't have it both ways, in other words, you can't say, "Bittorrent is the most efficient protocol man ever created, but only if you don't use it on 4g."

On the other hand, multicast with fountain codes (e.g. RaptorQ) sure seem to be a super efficient approach that might work everywhere.

[AnthonyMouse]

> I can make a very low latency connection but only delivers at most 1 packet every 10 seconds. It responds in 5 milliseconds say, but it won't let you have another packet for 10 seconds.

That's fine. So you connect to that peer and get one packet every ten seconds. You also connect to the other 1000 peers with the lowest latency, some of which are likely to be faster and in any event will be fast in aggregate, and none of which require traversing intercontinental fiber.