What is yespower?
yespower is a proof-of-work (PoW) focused fork of yescrypt. While
yescrypt is a password-based key derivation function (KDF) and password
hashing scheme, and thus is meant for processing passwords, yespower is
meant for processing trial inputs such as block headers (including
nonces) in PoW-based blockchains.
On its own, yespower isn't a complete proof-of-work system. Rather, in
the blockchain use case, yespower's return value is meant to be checked
for being numerically no greater than the blockchain's current target
(which is related to mining difficulty) or else the proof attempt
(yespower invocation) is to be repeated (with a different nonce) until
the condition is finally met (allowing a new block to be mined). This
process isn't specific to yespower and isn't part of yespower itself
(rather, it is similar in many PoW-based blockchains and is to be
defined and implemented externally to yespower) and thus isn't described
in here any further.
Why or why not yespower?
Different proof-of-work schemes in existence vary in many aspects,
including in friendliness to different types of hardware. There's
demand for all sorts of hardware (un)friendliness in those - for
different use cases and by different communities.
yespower in particular is designed to be CPU-friendly, GPU-unfriendly,
and FPGA/ASIC-neutral. In other words, it's meant to be relatively
efficient to compute on current CPUs and relatively inefficient on
current GPUs. Unfortunately, being GPU-unfriendly also means that
eventual FPGA and ASIC implementations will only compete with CPUs, and
at least ASICs will win over the CPUs (FPGAs might not because of this
market's peculiarities - large FPGAs are even more "over-priced" than
large CPUs are), albeit by far not to the extent they did e.g. for
Bitcoin and Litecoin.
There's a lot of talk about "ASIC resistance". What is (or should be)
meant by that is limiting the advantage of specialized ASICs. While
limiting the advantage at KDF to e.g. 10x and at password hashing to
e.g. 100x (talking orders of magnitude here, in whatever terms) may be
considered "ASIC resistant" (as compared to e.g. 100,000x we'd have
without trying), similar improvement factors are practically not "ASIC
resistant" for cryptocurrency mining where they can make all the
difference between CPU mining being profitable and not. There might
also exist in-between PoW use cases where moderate ASIC advantage is OK,
such as with non-cryptocurrency and/or private/permissioned blockchains.
Thus, current yespower may be considered either a short-term choice
(valid until one of its uses provides sufficient perceived incentive to
likely result in specialized ASICs) or a deliberate choice of a pro-CPU,
anti-GPU, moderately-pro-ASIC PoW scheme. It is also possible to
respond to known improvements in future GPUs/implementations and/or to
ASICs with new versions of yespower that users would need to switch to.
yespower versions.
yespower includes optimized and specialized re-implementation of the
obsolete yescrypt 0.5 (based off its first submission to Password
Hashing Competition back in 2014) now re-released as yespower 0.5, and
brand new proof-of-work specific variation known as yespower 1.0.
yespower 0.5 is intended as a compatible upgrade for cryptocurrencies
that already use yescrypt 0.5 (providing a few percent speedup), and
yespower 1.0 may be used as a further upgrade or a new choice of PoW by
those and other cryptocurrencies and other projects.
There are many significant differences between yespower 0.5 and 1.0
under the hood, but the main user visible difference is yespower 1.0
greatly improving on GPU-unfriendliness in light of improvements seen in
modern GPUs (up to and including NVIDIA Volta) and GPU implementations
of yescrypt 0.5. This is achieved mostly through greater use of CPUs'
L2 cache.
The version of algorithm to use is requested through parameters,
allowing for both algorithms to co-exist in client and miner
implementations (such as in preparation for a cryptocurrency hard-fork
and/or supporting multiple cryptocurrencies in one program).
Parameter selection.
For new uses of yespower, set the requested version to the highest
supported, and set N*r to the highest you can reasonably afford in terms
of proof verification time (which might in turn be determined by desired
share rate per mining pool server), using one of the following options:
1 MiB: N = 1024, r = 8
2 MiB: N = 2048, r = 8
4 MiB: N = 1024, r = 32
8 MiB: N = 2048, r = 32
16 MiB: N = 4096, r = 32
and so on for higher N keeping r=32.
You may also set the personalization string to your liking, but that is
not required (you can set its pointer to NULL and its length to 0). Its
support is provided mostly for compatibility with existing modifications
of yescrypt 0.5.
Performance.
Please refer to PERFORMANCE for some benchmarks and performance tuning.
How to test yespower for proper operation.
On a Unix-like system, invoke "make check". This will build and run a
program called "tests", and check its output against the supplied file
TESTS-OK. If everything matches, the final line of output should be the
word "PASSED".
We do most of our testing on Linux systems with gcc. The supplied
Makefile assumes that you use gcc.
Alternate code versions and make targets.
Two implementations of yespower are included: reference and optimized.
By default, the optimized implementation is built. Internally, the
optimized implementation uses conditional compilation to choose between
usage of various SIMD instruction sets where supported and scalar code.
The reference implementation is unoptimized and is very slow, but it has
simpler and shorter source code. Its purpose is to provide a simple
human- and machine-readable specification that implementations intended
for actual use should be tested against. It is deliberately mostly not
optimized, and it is not meant to be used in production.
Similarly to "make check", there's "make check-ref" to build and test
the reference implementation. There's also "make ref" to build the
reference implementation and have the "benchmark" program use it.
"make clean" may need to be run between making different builds.
How to integrate yespower in a program.
Although yespower.h provides several functions, chances are that you
will only need to use yespower_tls(). Please see the comment on this
function in yespower.h and its example usage in tests.c and benchmark.c,
including parameter sets requesting yescrypt 0.5 as used by certain
existing cryptocurrencies.
To integrate yespower in an altcoin based on Bitcoin Core, you might
invoke yespower_tls() from either a maybe-new (depending on where you
fork from) CBlockHeader::GetPoWHash() (and invoke that where PoW is
needed like e.g. Litecoin does for scrypt) or CBlockHeader::GetHash()
(and implement caching for its return value like e.g. YACoin does for
scrypt). Further detail on this (generating new genesis blocks, etc.)
is not yespower-specific and thus is not provided here. Just like (and
even more so than) yespower itself, this guidance is provided as-is and
without guarantee of being correct and safe to follow. You're supposed
to know what you're doing.
Credits.
scrypt has been designed by Colin Percival. yescrypt and yespower have
been designed by Solar Designer building upon scrypt.
The following other people and projects have also indirectly helped make
yespower what it is:
- Bill Cox
- Rich Felker
- Anthony Ferrara
- Christian Forler
- Taylor Hornby
- Dmitry Khovratovich
- Samuel Neves
- Marcos Simplicio
- Ken T Takusagawa
- Jakob Wenzel
- Christian Winnerlein
- DARPA Cyber Fast Track
- Password Hashing Competition
Contact info.
First, please check the yespower homepage for new versions, etc.:
http://www.openwall.com/yespower/
If you have anything valuable to add or a non-trivial question to ask,
you may contact the maintainer of yespower at:
Solar Designer <solar at openwall.com>