I will just repost some of my findings from the multipool thread from yesterday.
As I understand it, the algorithm works like this:
Next Diff = (Average Diff over the last N blocks) * (150 seconds) * N / (Actual Seconds for last N blocks)
where N is a number between 144 and 4032.
The "clever" part is that N gets smaller when the diff changes quickly and N gets larger when the diff is stable. The idea would be that when there is a quick change in hash rate, it reacts quickly. When the hash rate is stable, it reacts slowly.
The problem is the first part of the algorithm has momentum. This is one of the things that caused so much trouble with the transition at block 42,000. Even after the MP switched off around block 42130, the difficulty kept climbing for about 70 more blocks (which took 2 days).
For example, if N is 144 and the coin-hoppers mine a fast 50 coins and switch off the coin, those coins will still be counted in the average used for the next difficulty while the faithful have to grind on 94 coins themselves.
The change in direction will also increase N, causing the permitted change to be smaller, and allowing those 50 fast-mined coins to linger even longer, forcing the faithful miners to spend even more time trying to create even more blocks necessary to bring difficulty back down.
The idea was clever, but not practically applied. If instead of using block counts for PastBlockMin and PastBlockMax, he had used Time (i.e. PastSecondsMin and PastSecondsMax), then used blocks/hour instead of hours (seconds) per block for the calculation. That way if the coin slows down to 2-4 blocks per hour, it doesn't take days of grinding to get the 50-100 blocks needed to push the fast blocks out of the window of consideration.
Hopefully, I am wrong and it stays profitable to mine for the faithful and keeps producing blocks. Unfortunately, I think that every time it gets marginally profitable to mine, MP, MC and CC will slap it down and create problem like it had with block 42,000 where transactions don't confirm and 2 guys are stuck mining it for days waiting for it to recover, only to get slapped down and have to repeat the cycle.
To give some real numbers, last night, when middlecoin, multipool, and coinchoose were off mec, it took 16 hours to generate 60 blocks, about half of what would be required to push fast block out of the minimum 144 block consideration window. That was with a descent profitability. If profitability drops to a very undesirable level, like around block 42159-42219, mining 60 blocks took 3.5 days.
Hey raven,
Also curious to its real-world performance.
Will replace it if fails to serve.
We will find out shortly.
Very exciting times!
See you soon!
Dr Kimoto
Chan