In Bitcoin the proof of work is the same as was defined in a previous generation of virtual currency called hashcash. In order to generate Bitcoins you have to solve the mathematical problem set out in the definition of the proof of work - a process that theoretically everyone trying to solve the problem is equally likely to solve. It is exactly this process that Bitcoin Miners undertake to earn their reward of Bitcoins.
The problem for Bitcoin Miners, and some would say Bitcoin as a currency, is that the original definition of Bitcoin means that the reward for solving the problem progressively decreases over time so that eventually the true economic costs of mining Bitcoins means that it will no longer we worth doing so for the reward it attracts. This means that there is a likely to be a limit to the number of Bitcoins ever generated (estimated to be around 21 million).
Even today the infrastructure required to undertake Bitcoin Mining and see any meaningful return is serious business. It's no longer enthusiasts with lashed up Raspberry PIs running off some glowing 13 amp plug. It's large warehouses in areas where electricity is cheap(er) and cooling costs lower.
|A modern Bitcoin Mining operation|
Not surprising then that there has been research into how one might increase the profits from Bitcoin Mining. Much of the work has been on how to improve the hash rates of hardware or similar methods of improving the speed at which existing systems can calculate the key parts of the proof of work problem.
Then the researchers from the University of Illinois took a different approach. They looked at the algorithms behind the proof of work problem. They recognised that the problem was based on an algorithm that was what they term "...embarrassingly parallel and probabilistic in nature..."
The researchers took a different approach to computing the solution. It is a technique known as Approximate Computing (a technique I first became aware of in papers such as this from Universities of Alberta and Austin). It is a technique that is particularly useful in algorithms that are not particularly sensitive to errors and it works on the basis that an answer is good enough if it is just good enough. The original paper I read puts it thus "By relaxing the need for fully precise or completely deterministic operations, approximate computing techniques allow substantially improved energy efficiency."
And it appears to pay large dividends when applied to Bitcoin Mining. The results reported in the paper just released show that profits can be increased by 30% for a Bitcoin Mining operation if they adopt approximate computing techniques.
As the cost of mining Bitcoins is reaching a point where every penny of cost counts, this latest finding could allow those miners who adopt it to steal a march on their competitors. The question it poses though is, is it fair? After all the original concept was that the problem was equally difficult, for those with equal computing power, to solve.
Personally I don't think this is new finding is any different to optimising hardware for Bitcoin Mining (which as been happening for some time now). All that is happening now is that computer science rather than electronic engineering has become involved. I see it as analogous to drilling for oil - when oil was found in the North Sea a whole new range of technologies were developed because it had become economic to do so. People didn't say that it was too difficult to extract or that it was any less valuable once extracted.
I think Bitcoins Mining will be subject to technical advances in the same way as mining for natural resources that become increasingly scarce. Maybe that was what Satoshi had in mind all along. Maybe one day we'll be able to ask him/her/them?