But wait – you can become your own validator! [...]
The consensus system ensures there is only one ledger – the most accepted one.
If you run your own validator, you must connect with OpenCoin Inc servers, or you
will be building a different ledger. In practice, it is not possible to
“dethrone” OpenCoin Inc as you have to co-operate with OpenCoin Inc.
Second, I'll point out that while minimizing the number of validators may be attractive, it also makes the currency much more vulnerable to top-down control, via governments, corporations, etc.Last, if you're reading this, I'd suggest that you read up a bit about Ripple on Bitcoin Forum before you make conclusions about the currency - there are a lot of interesting discussions about its viability (like this one: https://bitcointalk.org/index.php?topic=146964.0).
I don't see why minimizing the number of validators is attractive. We want as many as possible precisely because that ensures that administrative control can't be seized. It's just like mining in Bitcoin -- you want it distributed into the hands of as many people as possible as evenly as possible. We are working on making this happen now. Obviously, open sourcing the server was a necessary stop to broadening the validators.
So the net result is that, since everyone - including the exchanges - trusts the OpenCoin-run nodes, the only safe thing to do is wait for OpenCoin's servers to come to a consensus amongst themselves and then accept their ledger updates in their entirety. There is no way for anyone outside of OpenCoin Inc to influence what gets included in the ledger except if you allow it.
It is true that today we control the majority of validators that other important servers trust. We're working now on increasing the number of validators because that will improve the reliability and robustness of the network. We absolutely do not want people to have to trust us, or even think they have to trust us, so this is a real priority for us. Open sourcing the server was, obviously, a step in this direction.
Please see the thread linked to above. There were several trust topologies given than are obviously non-convergent in your model and you appeared to have no answer for them.
Effectively you have ripple servers which must have central selection/control (or the system is not guaranteed to converge), and they can set balances to whatever they want. ... and then clients that simply believe the majority of the servers they are pointed at.
IOW, attacker-controlled Ripple nodes can outnumber legitimate Ripple nodes, therefore legit Ripple nodes are forced to accept transactions broadcasted by attacker-controlled nodes.
Ripple does have only a small number of server nodes today, but that should be growing over time. A node cannot freely set balances. Each node signs each ledger and any change to a ledger entry (such as a balance) must be accompanied by a signed transaction justifying that change or other nodes will reject it.
Nodes are in fact called "validators" and they validate each ledger to ensure that any changes are justified by transactions. They sign these ledgers every few seconds.
As for the peering needing to be centrally controlled, I assume you're referring to each node's set of validators. While that does need to be sanely managed, it doesn't need to be centrally controlled. A number of organizations can publish lists of validators they believe to be reliable. The algorithm is very tolerant of things like minimal overlap or bad apples. Every honest node wants to agree with every other honest node, and dishonesty is impossible to hide, so the problem is not that difficult.
(I'm one of the architects of the Ripple network.)