README
¶
Proof of Work Algorithms
Overview
Note: This library is still in active development and is subject to breaking changes
Though there is a wide variety of Proof of Work algorithms, finding the technical details for the implementations is quite a task. Both Kawpow and Firopow are variations off of ProgPow, though finding the exact differences is no easy task. This is meant to be a unified library to make the implementation of existing Proof of Work algorithms easier.
All DAG-based algorithms only implement a light DAG, which is sufficient for verification
but not full nodes or miners. For the DAG-based algorithms and verthash, data is cached in ~/.powcache.
Ethash will generally be between 40-80Mb per epoch (and generally 3 caches are stored), but verthash
requires a strict 1.2Gb, so be careful if you're using verthash in memory. At the time of writing, running
make test will throw about 3.3Gb of data into ~/.powcache due to the variety and breadth of tests.
Algorithms
| Algorithm | DAG | Supported |
|---|---|---|
| Ethash | yes | yes |
| Etchash | yes | yes |
| Kawpow | yes | yes |
| Firopow | yes | yes |
| Octopus | yes | yes |
| Verthash | yes | yes |
| Equihash | no | yes |
| Autolykos2 | no | yes |
| Cuckoo Cycle | no | yes |
| Eaglesong | no | yes |
| BeamHashIII | no | no |
Things to Note
- There is no validation on hash input sizes - generally it is 32 bytes but it can vary by algorithm. It absolutely can panic if the wrong input size is used. I probably should be stricter about this, or at least define what the expectations are, but for now the best way to check is in the tests.
- Most of these algorithms are partially optimized, though there surely could be some improvements. However, that is not the goal since these have never been intended to be used for miners. All of these algorithms far surpass a reasonable threshold for performance and I have no intention of hypertuning them.
- Cuckoo Cycle is built specifically for Aeternity. There is a modification of the
sipnodefunction in the current version of tromp's cuckoo algorithms that Aeternity does not use (a Nicehash dev gives more details here). It wouldn't be hard to implement other Cuckoo Cycle algorithms (cuckatoo, cuckaroo), there just isn't really a need at this point since Grin is fairly annoying. BlockCypher implements the other algorithms here. - All non-DAG algorithms are less organized than I would like, they'll probably be overhauled at some point for a more coherent general standard.
- All testing is done on linux, windows support is hazy at best.
- The library assumes the host architecture is little-endian, I'm fairly confident big-endian architectures will not function properly.
- The base ProgPow implementation ("ProgPow094") exists in the
internal/progpowpackage. - One day I'll implement BeamHashIII, it is just a slight modification of Equihash (I think 150_5?). Other than that, no other algorithms are planned - there is a cryptonight and a randomx implementation in Go, though these aren't really of interest. Of course, this can change if new algorithms become popular.
Roadmap
Most profitable Proof of Work chains nowdays use some sort of DAG, and that is almost always the Ethash DAG. There are then layers on top of the Ethash DAG like ProgPow, which can be considered both a Proof of Work algorithm and a class of algorithms (Kawpow, Firopow). There isn't really a good way to organize all of these into a given structure but even if there was, a new algorithm could appear tomorrow and break that structure. powkit takes the approach of allowing most parameters to be varied and will implement new algorithms on an as-needed basis.
Currently, though powkit is used in production internally, it probably isn't a good idea to use yourself. The API is still in flux and each minor version will probably be breaking. Once we do a v1.0.0 release, the structure will probably be pretty set in stone.
Directories