Interesting paper, you must do a lot of reading on this stuff!
The way these guys did it would be the best way to do it if we wanted to do an exhaustive search of the key space. However, for bitcoin addresses, each feasible pattern has a ridiculously large number of matching keys. Even searching for a complete address will have, on average, 2^96 matching keys. So, there's almost no point in doing an exhaustive search.
The way I see it, there is no performance benefit in supporting multiple GPU's from the same instance. It's just aesthetic.
That is the idea, it is mostly aesthetic. Instead of having to run oclvanitygen once per GPU, you'll be able to run it once per system, and tell it to use all of your GPUs.
Organized clustering across systems will be more complicated. One way to do this, in the near future, that follows the model of bitcoin mining, will be:
- Run oclvanityminer on each system in the cluster.
- Run a private bounty server, or the equivalent of a mining proxy, and configure your oclvanityminers to connect to it.
There isn't a private bounty server, at least not yet, but all it would need to do is to support the
vanity pool protocol.
