In May 2014 someone wrote:
p2p is no panacea, it doesn't scale
I believe it could. Even if requiring super aggregating nodes of some sort. Layers of service of the whole DHT space. More research is surely required.
It is not possible to have fast p2p unless: - Cable networks collaborate by increasing bandwidth 7 to 8 times
My references to scale were not intended to be about... bulk bandwidth across such networks (for example, right now, I2P and Tor are doing well enough to see very low quality video between their hidden nodes if you get a lucky path, and well enough for moving large files around in non realtime). ie: the nodes have bandwidth available. But about scaling the node (user) count from millions to billions... No device (or its ethernet) will be able to manage a 10 billion entry DHT with a handful of keys, addresses and flags per entry. But if you break it up into some many clusters/hiers/roles of smaller DHT's, each knowing how to route queries, sort, and pass entries around, it might work. Once you've assembled your multihop path from querying the DHT for nodes, actual data transfer rates should remain similar. (Provided the network clients know to reserve bandwidth mod the network average hop count, by throttling the users usage at their own node). It would be nice to check some numbers on this for the list. Is there a wiki or paper repository that discusses plausibly reachable DHT sizes, time needed for DHT ops to resolve, and management schemes for such clusters/hiers/roles? [aside: This everyone online, big DHT, end-to-end reachable model mirrors the internet today as a general purpose tool. Perhaps sufficient for many rather sensitive tasks. When the purpose is narrowed, other models may apply. For messaging (as is the subject), everyone still needs a unique address. And since msg delivery/pickup must be reliable, there is a question of redundancy needed to avoid random msg loss. Which may turn you away from store-forward, mixes, and unconscious central storage, etc... towards everyone online, contact them directly over a path or retry later. Today it seems that general purpose may be better researched and easier than more exotic means. Question is, is GP sufficient, after applying any recent GP tech post I2P and Tor's designs? ie: Some say timing attacks may be mitigated by fixed packet lengths and adding chaff over links as cover.]