On Fri, 2011-08-05 at 01:52 -0400, Jeff Garzik wrote:
> Yes, that is correct. Bitcoin resends wallet transactions with zero
> confirmations, and both sent and received transactions fall within the
> "wallet tx" superset.
>
> TBH I had forgotten about the resend on the receiver side, though.
> It, of course, makes plenty of sense in the context of importing
> transactions from foreign sources, e.g. receiving transactions via a
> USB flash drive.
Could every node do the resends? Alternatively, could we implement a TOR
like tunneling system just for the first leg of the transactions
(overkill?). Then again, maybe just a TOR gateway if that's desired.
> > Drawok's suggestion about using UDP packets with spoofed sender addresses is
> > interesting, as UDP has another advantage; you can open up an "inbound" UDP
> > port on almost any NAT router without any UPNP magic: just send out an UDP
> > packet, the router will wait a certain time for answers (on a mapped port
> > number) and relay these back.
This is a nice idea but sounds rather unreliable.
> Well, it -is- possible to implement TCP over UDP <grin> The TCP
> connection sequence over UDP helps to work against spoofing, while UDP
> helps to open an inbound UDP port as you describe.
There's already an implementation of this, called UTP. If we do decide
that using UDP is worthwhile, this library is probably better than
implementing something ourselves.
- Joel