Quote from: ThePiachu on October 20, 2012, 03:35:15 PM

I'm currently working on some metrics for calculating how profitable the pool is for the miners. My initial ideas are:

Variables:
Bitcoin Difficulty d
Miner Mining Speed m
Block Reward r

Miner Key Generation Rate k
Vanity Address Bounty b
Vanity Address Complexity (aka "difficulty", but I don't want to confuse the two) c

Expected Block Reward: (m*r)/(d*2^32)
Expected Vanity Reward: (k*b)/c

Compare the two and see which is higher.

However, if I want to present this more neatly in one equation, it would probably be like this:

Ratio: (b*2^32)/c * d/r * k/m

The right part is constant for a given miner, the left part is constant for a given pattern, and the middle part is just a simple ratio of difficulty and block reward. This equation is for a single address. If the ratio is higher than 1, you ought to mine for addresses, otherwise - mine for blocks.

I think I will call the left part of this equation Lavishness. To determine if mining at the Pool is more profitable than mining, one would all various Lavishnesses and put them in together into the equation. For multiplicative mining - one would take the biggest Lavishness sum of patterns sharing a single public key, and for additive mining - add all the Lavishnesses of all patterns (but also keep in mind different key generation rate for this method).

So what do you think, is this a good metric?

Yes, this is exactly what I've been lobbying for.

In fact, I went ahead and did this yesterday (how I got the numbers in the post above).

In fact, I plugged these formulas into a script that grabs the relevant quantities from blockchain.info and parses your available work list to calculate them at any time. Then, it feeds the results into munin so that I can see a graph over time, so I can see some historical data on this:

As you can see, the "value" of address mining has decreased a little bit since the graph started because some work was solved. When the bitcoin difficulty or block reward changes, the green line will change, too.

Using your variable definitions, My graph is made with a ratio for m/k = 400/30 (which is true for my 5870). Could we collect a bunch of m/k values for a lot of different hardware and average them? If that ratio is pretty much constant for most GPUs, this value would be meaningful for most users (right now, it is only meaningful for a 5870).

If you'd like to help this effort, please post your performance figures for your hardware (GPU model and hash/s and key/s).