Proof of Work
Proof of work is a consensus mechanism which allows anyone to extend the blockchain, by providing a piece of data that is both difficult to produce and easy to verify (according to set parameters). Proof of work also serves as the basis for the security and distribution of the coin.
Grin has an average block time of 60 seconds and employs Cuckoo Cycle1, a memory bound proof of work algorithm, or more specifically, a variation of it named Cuckatoo that is meant to simplify ASICs.
The algorithm finds length-42 cycles in a bipartite graph. The current (and final) grin PoW is Cuckatoo32+, in which a graph must have at least 232 + 232 nodes. For a comprehensive introduction, read here.
Q: Why use a memory-hard proof of work?
A: The point is that chips dominated by memory have rather different characteristics from computational chips; they run much cooler, dissipating less heat per unit of area. This shifts the mining cost from mostly OPEX (electricity) to mostly CAPEX (hardware cost), which delays obsolescence and allows mining in places with higher electricity costs.
Grin has two PoW algorithms:
- Cuckatoo32+, The primary algorithm, meant to be ASIC-friendly. Begins at 10% and evolves to 100% over 2 years.
- Cuckaroo29, The secondary algorithm, aimed to be ASIC-resistant. Begins at 90% and decreases to 0% after 2 years.
During the first 2 years of grin, the algorithms gradually balance themselves to satisfy the current ratio, starting from 90/10 and eventually becoming 0/100, such that the only proof of work remaining after those initial 2 years would be Cuckatoo32+. This period includes scheduled hardforks every 6 months, in which Cuckaroo29 is modified in order to interrupt possible ASIC development for it.
The above simplifies a bit, since in practice Cuckatoo31+ was initially the primary PoW but was phased out completely after 18 months, for reasons we won’t discuss here. You can read about it on this post.
ASICs are special pieces of hardware especially designed to do a specific task as quickly and efficiently as possible. In our context, they solve PoW a algorithm at a much faster rate than general purpose hardware, such as consumer GPUs. The common arguments against ASICs are about how they raise the barrier to entry for mining, and the centralization that may occur when a single ASIC manufacturer has complete market dominance.
These issues are mostly solved once a mature, competitive market for ASICs appears. However, as became apparent with bitcoin, this natural process may take time. Grin’s Cuckatoo32+ simplifies ASIC design in hope to reduce the time until they become widely distributed and accessible.
The choice of upholding some ASIC-resistance in the first 2 years was made to ensure fair initial distribution, in which no single party has a disproportional mining advantage at launch, before the market becomes populated and competitive enough.
Let’s see why encouraging ASIC development might be beneficial2 in the first place:
The security of a chain depends both on the amount of capital allocated to mine it (CAPEX), and on the cost of electricity & operation (OPEX). But the CAPEX is only relevant if the mining hardware's main function is to mine our specific chain and is mostly useless otherwise.
We should want to avoid having the chain prone to attack by large hardware operators who own no stake in its success (no skin-in-the-game), since they can attack it and divert their hardware to other uses without incurring any loss to their capital.
Therefore, to achieve optimal security, two conditions must be met:
- ASICs exist and perform significantly better than general-purpose hardware.
- Our chain has the largest economic value for its specific ASIC. If the same PoW algorithm is used by a bigger chain or serves another purpose, we encounter the same issue as described earlier.
There’s increasingly more evidence that ASICs are inevitable, as dedicated hardware will always have ways in which it can improve upon general purpose hardware. While it is possible to make ASIC manufacturing more difficult, over a long period it is likely to end up a centralizing force in itself, as it makes the chain vulnerable to secret ASIC operations3.