If I'm trying to transact in Dagcoin, it doesn't matter to me that maybe 100,000 nodes could potentially have my parent transaction, the only thing that matters to me is finding the node which does have it, and that's where my new transaction is going to go. The next transaction, which relies upon mine, is similarly going to go to that node, and so on. No one wants their transactions left out on a loose strand.
On the flipside, nodes want to get as many transactions as they possibly can because that's presumably where they'll get their revenue in Dagcoin.
So there definitely is an incentive on both sides to centralize the network.
Separately, the amount of bandwidth necessary to store a transaction at VISA is orders of magnitude less than what it costs to propagate a transaction across a network. So if we were looking at you're example, and say doing 2000KB/s for an average of 2000TPS, we would need at least 7000 times that bandwidth to comfortably propagate over 7000 nodes or 218.75MB/s (1750Mb/s).
Looks like we are talking about different DAG-coins. My version of DAG doesn't care about network topology.
Separately, the amount of bandwidth necessary to store a transaction at VISA is orders of magnitude less than what it costs to propagate a transaction across a network. So if we were looking at you're example, and say doing 2000KB/s for an average of 2000TPS, we would need at least 7000 times that bandwidth to comfortably propagate over 7000 nodes or 218.75MB/s (1750Mb/s).
Let's check your numbers. IP header size = 40 bytes, UDP header size = 8 bytes, transaction size = 1024 bytes. 40 + 8 + 1024 = 1072. Small world topology with 3 neighbors (this is enough to effectively propagate data) requires to send 3 times of that, which is 25'728 bits. For 100 Mbps we get 3'886 TPS. And this is without multicast enabled on routers between nodes.