Bump.
And I don't want to read the code to understand it.
AFAIK the server source isn't open yet anyway.
I will list my (mis-)conceptions to make it easier to debunk for whoever explains it.
I have been using ripplepay (the centralized predecessor) for a long time and how it works is pretty straightforward. The system finds least resistant trust path(s) to destination and shifts IOU's accordingly. If one path is not enough, it would find multiple paths until the required amount is achieved. I'm a little fuzzy about the details though, mainly whether it's possible to exhaust all possible paths and proportionally distribute IOU's and how length vs. trust is balanced. If you put the complexity aside, it's not hard to get with a centralized system.
There has been some work I came across in time that tried to do this in a fully distributed manner. It's clear that it can't reasonably be done if nodes and identities are the same, but it can actually be done with a distributed ledger a la Bitcoin.
AFAICT Ripple (with a capital R this time) hopes to minimize the trust requirement by emphasizing gateways. So aside from using it just like ripplepay (but in a distributed fashion), you can just use it to transfer money between gateways. Since gateways are supposed to be trustworthy institutions in their own right, users can just deposit real money to one of them (e.g. bank wire) and receive IOU's. When they send these IOU's to another gateway, payment path goes through the original gateway to the destination. Effectively, the money you've depisited at gateway1 goes to gateway2 and gateway1 now owes money to gateway2. To me, it seems like generic, standardized coupon codes with architectural support for account limits.
Again, AFAICT, this is just how they plan to accelerate adoption and has nothing to do with the fundamentals. A gateway is no different than any other node.
The public database not only holds the trust paths and their accounts, they hold exchange offers. So, you can announce that you are willing to exchange x of A with y of B, and it goes in this database. The system then makes the transfer automatically when someone accepts the offer. This is my understanding of the distributed exchange feature.
Presumably, the nodes do these calculations locally and then share with the network. When a consensus is reached, it goes permanently in the ledger.
I only superficially checked out Ripple, this is all "how I would do it" kind if guess on my part, so feel free to crush my self confidence.
