Would Tor work if it became very large?

Question

I wonder whether there would be any problems if Tor grew to serve a large portion (e.g. 5 or 30 %) of internet (web) usage?

As I understand it Tor is helped by new nodes at the moment – but is there any limit when its usability or safety would decrease?

Answer 1

Caveat: not a complete answer, but still a start:

There probably is a limit, given Tor's current design. For instance, right now it's assumed that every relay can and does talk directly to every other relay. Once you have many tens of thousands of relays that just won't work anymore.

For now, Tor has kept ahead of the curve. However, this requires constant updates, discussion, and reflection. There is a discussion from a couple years back in the tor specification repository.

Answer 2

Another bottleneck is the distribution of the network consensus. To maintain anonymity, every client must have the full consensus for relay selection. As more relays and users join the network, the distribution gets more and more expensive. After all, every relay adds another entry to the consensus and every user will periodically want to download the consensus.

This paper gives a good overview of the problem: http://freehaven.net/anonbib/#ccs09-torsk

Answer 3

The whole Tor network advertises 6 GiB/s according to the metrics page. DE-CIX statistics say they have an average bandwidth of 200 GiB/s. So if you assume that web traffic is roughly 20% of all traffic, we talk about 40 GiB/s. Numbers between 5 and 30% are 2 to 12 GiB/s of traffic in the DE-CIX case.

So if those estimation is correct, Tor should be able to handle (DE-CIX) web traffic right now. It would need some fast relays, but not enormous amounts of relays. I assume the numbers don't grow that much on a worldwide base. So with a fair number of fast relays web traffic should work.

However when when talk about general Internet traffic (BitTorrent alone is estimated to account for >40% of internet traffic), you will needs ten of thousands, if not millions of relays. A client will have to download all those information at the beginning. I would estimate that this file can easily grow to some hundreds of MB. Furthermore, as Peter wrote, every relay talks to each other which simply doesn't work with so many relays.