EIP 1485: TEthashV1
| Author | trustfarm (KT Ahn - 안씨아저씨), trustfarm |
|---|---|
| Discussions-To | https://ethereum-magicians.org/t/anti-eth-asic-mining-eip-1488-pr/1807 |
| Status | Draft |
| Type | Standards Track |
| Category | Core |
| Created | 2018-11-01 |
Simple Summary
This EIP modifies ethash in order to break ASIC miners specialized for the current ethash mining algorithm.
Abstract
This EIP pursue “obsolete current ASIC miners” by modifying PoW algorithm in a very low risk manner and update to latest hash algorithm from deprecated FNV Hash algorithms.
Following TEthashV1 algorithm suggests safe transition of PoW algorithms and secure the FNV Algorithm in MIX Parts.
Motivation
Provide original Ethash proof of work verification with minimal set of changes by updating FNV0 algorithm
Specification
1. Reference materials on ETHASH FNV0
Where FNV Applied on ETHASH
- In ETHASH , FNV Hash is used on
-
1) On data aggregation function, MIX parts.
-
Ethash Algorithm
Header + Nonce | Keccak | **[MIX 0]** --> **[DAG Page]** | | Mixing <--| ... | **[Mix 63]** | |-----> Mix64 [Process] ---> Mix Digest [32B]- FNV used in DAG Generation and Mixing for random access or DAG Page.
-
2. Current applied Ethash FNV hash implementation is deprecated now.
[FNV-0hash (deprecated)](https://en.wikipedia.org/wiki/Fowler%E2%80%93Noll%E2%80%93Vo_hash_function#FNV-0_hash(deprecated))
It is a simple way of hashing algorithm
hash = 0
for each byte_of_data to be hashed
hash = hash × FNV_prime
hash = hash XOR octet_of_data
return hash
When analysed FNV-0 , there’s very weak avalanche effect, when hash input changes on 1~2bits. refer FNV-Analysis reference section
We need to research and apply newer FNV hash or short message hash algorithm.
3. FNV1A hash algorithm description
Previous proposed algorithm based on FNV1 EIP-1355
There’s a implementation that looks like “Missing Offset Bias” at FNV1A.
Quotation of original algorithm FNV1A
use hash offset
FNV-1a hash
The FNV-1a hash differs from the FNV-1 hash by only the order in which the multiply and XOR is performed:[8][10]
hash = FNV_offset_basis
for each byte_of_data to be hashed
hash = hash XOR byte_of_data
hash = hash × FNV_prime
return hash
FNV_offset_basis and computation order change of xor and multiplication Makes one more xor and multiply computation, but more secure hash effects than FNV0. and make dispersion boundary condition (0, even number, ..) by using of Prime Number.
4. Real Implementation for FNV1A
Consider real computation resources, in TEthashV1 uses hash byte_of_data to 4bytes aligned data.
In TETHashV1, Adapts fully follow the FNV1A implementation.
- TETHASHV1 FNV1A implementation
Following are reference implementation of FNV1A adapted in TETHashV1.
// Reference Pseudo c/cpp implementation
#define FNV_PRIME 0x01000193U
#define FNV_OFFSET_BASIS 0x811c9dc5U
#define fnv1a(x, y) ((((FNV_OFFSET_BASIS^(x))*FNV_PRIME) ^ (y)) * FNV_PRIME)
#define fnv1a_reduce(a,b,c,d) (fnv1a(fnv1a(fnv1a(a, b), c), d))
Another Byte aligned implementation of FNV1A , call to FNV1c
#define FNV_PRIME 0x01000193U
#define FNV_OFFSET_BASIS 0x811c9dc5U
#define fnv1i(x) ( (( (( (( \
( ((FNV_OFFSET_BASIS)^( ((x)>>24)&0x000000ff )) * FNV_PRIME) \
^ (((x)>>16 )&0x000000ff)) * FNV_PRIME) \
^ (((x)>>8 )&0x000000ff)) * FNV_PRIME) \
^ (((x) )&0x000000ff)) * FNV_PRIME) \
)
#define fnv1c(x, y) ((fnv1i(x) ^ (y)) * FNV_PRIME)
5. FNV-Analysis
FNV Mix Algorithm Analysis for TEthashV1
How to test and analysis reference test code.
You can compile it with simple in terminal. No additional library needs,
gcc -o fnvtest fnvcltest.c
And You can execute it
fnvtest
F(00,00)::VEC(0, 0, ffffffff, 0):: FNV :00000000, DF=00000000(00) DS(00000000), FNV1 :00000000, DF=00000000(00) DS(00000000), FNV1a:117697cd, DF=117697cd(17) DS(117697cd), FNV1c:1210d00f, DF=127f8dbf(20) DS(11a1725f), F___RC=efe1b9c4, DF:efe1b9c4(19) , F1__RC=deb68dfe, DF:deb68dfe(22) , F1A_RC=99bad28b, DF:99bad28b(17) , F1C_RC=e29fa497, DF:e29fa497(18)
F(00,01)::VEC(0, 1, ffffffff, 0):: FNV :00000001, DF=00000001(01) DS(00000001), FNV1 :01000193, DF=01000193(06) DS(01000193), FNV1a:1076963a, DF=010001f7(09) DS(01000193), FNV1c:1110ce7c, DF=03001e73(11) DS(01000193), F___RC=fefffe6d, DF:111e47a9(14) , F1__RC=d9fd8597, DF:074b0869(12) , F1A_RC=72c287e0, DF:eb78556b(19) , F1C_RC=6b6991ef, DF:89f63578(17)
F(00,02)::VEC(0, 2, ffffffff, 0):: FNV :00000002, DF=00000003(02) DS(00000001), FNV1 :02000326, DF=030002b5(08) DS(01000193), FNV1a:0f7694a7, DF=1f00029d(11) DS(01000193), FNV1c:1410d335, DF=05001d49(09) DS(030004b9), F___RC=d8fd8404, DF:26027a69(13) , F1__RC=9b16d24c, DF:42eb57db(19) , F1A_RC=c17f0ecb, DF:b3bd892b(18) , F1C_RC=a5be8e78, DF:ced71f97(21)
F(00,03)::VEC(0, 3, ffffffff, 0):: FNV :00000003, DF=00000001(01) DS(00000001), FNV1 :030004b9, DF=0100079f(10) DS(01000193), FNV1a:0e769314, DF=010007b3(09) DS(01000193), FNV1c:1310d1a2, DF=07000297(09) DS(01000193), F___RC=b2fb099b, DF:6a068d9f(16) , F1__RC=5c301f01, DF:c726cd4d(17) , F1A_RC=94cf402e, DF:55b04ee5(16) , F1C_RC=aea1a025, DF:0b1f2e5d(17)
F(00,04)::VEC(0, 4, ffffffff, 0):: FNV :00000004, DF=00000007(03) DS(00000001), FNV1 :0400064c, DF=070002f5(10) DS(01000193), FNV1a:0d769181, DF=03000295(07) DS(01000193), FNV1c:0e10c9c3, DF=1d001861(09) DS(050007df), F___RC=8cf88f32, DF:3e0386a9(14) , F1__RC=1d496bb6, DF:417974b7(17) , F1A_RC=89401d59, DF:1d8f5d77(20) , F1C_RC=e4e96c7c, DF:4a48cc59(13)
F(00,05)::VEC(0, 5, ffffffff, 0):: FNV :00000005, DF=00000001(01) DS(00000001), FNV1 :050007df, DF=01000193(06) DS(01000193), FNV1a:0c768fee, DF=01001e6f(11) DS(01000193), FNV1c:0d10c830, DF=030001f3(09) DS(01000193), F___RC=66f614c9, DF:ea0e9bfb(20) , F1__RC=de62b86b, DF:c32bd3dd(19) , F1A_RC=346e222c, DF:bd2e3f75(21) , F1C_RC=502e5f82, DF:b4c733fe(20)
F(00,06)::VEC(0, 6, ffffffff, 0):: FNV :00000006, DF=00000003(02) DS(00000001), FNV1 :06000972, DF=03000ead(10) DS(01000193), FNV1a:0b768e5b, DF=070001b5(09) DS(01000193), FNV1c:1010cce9, DF=1d0004d9(10) DS(030004b9), F___RC=40f39a60, DF:26058ea9(13) , F1__RC=9f7c0520, DF:411ebd4b(16) , F1A_RC=b376a527, DF:8718870b(13) , F1C_RC=1241a9a4, DF:426ff626(17)
F(00,07)::VEC(0, 7, ffffffff, 0):: FNV :00000007, DF=00000001(01) DS(00000001), FNV1 :07000b05, DF=01000277(08) DS(01000193), FNV1a:0a768cc8, DF=01000293(06) DS(01000193), FNV1c:0f10cb56, DF=1f0007bf(15) DS(01000193), F___RC=1af11ff7, DF:5a028597(13) , F1__RC=609551d5, DF:ffe954f5(22) , F1A_RC=14293bea, DF:a75f9ecd(21) , F1C_RC=49d34bba, DF:5b92e21e(16)
F(00,08)::VEC(0, 8, ffffffff, 0):: FNV :00000008, DF=0000000f(04) DS(00000001), FNV1 :08000c98, DF=0f00079d(12) DS(01000193), FNV1a:09768b35, DF=030007fd(12) DS(01000193), FNV1c:1a10dca7, DF=150017f1(12) DS(0b001151), F___RC=f4eea58e, DF:ee1fba79(21) , F1__RC=21ae9e8a, DF:413bcf5f(19) , F1A_RC=eeebb7a5, DF:fac28c4f(17) , F1C_RC=7da04f47, DF:347304fd(16)
F(00,09)::VEC(0, 9, ffffffff, 0):: FNV :00000009, DF=00000001(01) DS(00000001), FNV1 :09000e2b, DF=010002b3(07) DS(01000193), FNV1a:087689a2, DF=01000297(07) DS(01000193), FNV1c:1910db14, DF=030007b3(10) DS(01000193), F___RC=ceec2b25, DF:3a028eab(14) , F1__RC=e2c7eb3f, DF:c36975b5(18) , F1A_RC=54e1aef8, DF:ba0a195d(15) , F1C_RC=d425e1af, DF:a985aee8(16)
F(00,0a)::VEC(0, a, ffffffff, 0):: FNV :0000000a, DF=00000003(02) DS(00000001), FNV1 :0a000fbe, DF=03000195(07) DS(01000193), FNV1a:0776880f, DF=0f0001ad(10) DS(01000193), FNV1c:1c10dfcd, DF=050004d9(08) DS(030004b9), F___RC=a8e9b0bc, DF:66059b99(15) , F1__RC=a3e137f4, DF:4126dccb(15) , F1A_RC=213fcd63, DF:75de639b(20) , F1C_RC=7e1d2751, DF:aa38c6fe(18)
F(00,0b)::VEC(0, b, ffffffff, 0):: FNV :0000000b, DF=00000001(01) DS(00000001), FNV1 :0b001151, DF=01001eef(12) DS(01000193), FNV1a:0676867c, DF=01000e73(09) DS(01000193), FNV1c:1b10de3a, DF=070001f7(11) DS(01000193), F___RC=82e73653, DF:2a0e86ef(16) , F1__RC=64fa84a9, DF:c71bb35d(19) , F1A_RC=5598ce46, DF:74a70325(14) , F1C_RC=6400c630, DF:1a1de161(14)
F(00,0c)::VEC(0, c, ffffffff, 0):: FNV :0000000c, DF=00000007(03) DS(00000001), FNV1 :0c0012e4, DF=070003b5(10) DS(01000193), FNV1a:057684e9, DF=03000295(07) DS(01000193), FNV1c:1610d65b, DF=0d000861(07) DS(050007df), F___RC=5ce4bbea, DF:de038db9(17) , F1__RC=2613d15e, DF:42e955f7(18) , F1A_RC=6a220ff1, DF:3fbac1b7(20) , F1C_RC=6e781da4, DF:0a78db94(15)
F(00,0d)::VEC(0, d, ffffffff, 0):: FNV :0000000d, DF=00000001(01) DS(00000001), FNV1 :0d001477, DF=01000693(07) DS(01000193), FNV1a:04768356, DF=010007bf(11) DS(01000193), FNV1c:1510d4c8, DF=03000293(07) DS(01000193), F___RC=36e24181, DF:6a06fa6b(17) , F1__RC=e72d1e13, DF:c13ecf4d(18) , F1A_RC=168d4944, DF:7caf46b5(19) , F1C_RC=65bbcfa1, DF:0bc3d205(13)
F(00,0e)::VEC(0, e, ffffffff, 0):: FNV :0000000e, DF=00000003(02) DS(00000001), FNV1 :0e00160a, DF=0300027d(09) DS(01000193), FNV1a:037681c3, DF=07000295(08) DS(01000193), FNV1c:1810d981, DF=0d000d49(09) DS(030004b9), F___RC=10dfc718, DF:263d8699(15) , F1__RC=a8466ac8, DF:4f6b74db(20) , F1A_RC=93e667bf, DF:856b2efb(19) , F1C_RC=76f80ee3, DF:1343c142(11)
F(00,0f)::VEC(0, f, ffffffff, 0):: FNV :0000000f, DF=00000001(01) DS(00000001), FNV1 :0f00179d, DF=01000197(07) DS(01000193), FNV1a:02768030, DF=010001f3(08) DS(01000193), FNV1c:1710d7ee, DF=0f000e6f(13) DS(01000193), F___RC=eadd4caf, DF:fa028bb7(17) , F1__RC=695fb77d, DF:c119ddb5(17) , F1A_RC=0f485682, DF:9cae313d(17) , F1C_RC=3667e8dc, DF:409fe63f(18)
F(00,10)::VEC(0, 10, ffffffff, 0):: FNV :00000010, DF=0000001f(05) DS(00000001), FNV1 :10001930, DF=1f000ead(13) DS(01000193), FNV1a:01767e9d, DF=0300fead(14) DS(01000193), FNV1c:0210b6df, DF=15006131(09) DS(1500210f), F___RC=c4dad246, DF:2e079ee9(17) , F1__RC=2a790432, DF:4326b34f(16) , F1A_RC=d10adebd, DF:de42883f(16) , F1C_RC=1ce48e12, DF:2a8366ce(15)
F(00,01) : is input x,y
VEC(0, 1, ffffffff, 0) : is fnv_reduce input vector (a,b,c,d)
FNV :00000001, DF=00000001(01) DS(00000001) :
FNV(00,01)result is 00000001 ,DF: is changed bitcounts, compared with previous outputs, in this case prev[00,00] current[00,01] input is 1bit changed, and output result 1bit changed.DS: is distances of previous result and current result , ABS(prev_fnvresult,current_fnvresult).
** Basically, DF is higher is best on hash algorithm.
F___RC=fefffe6d, DF:111e47a9(14) : fnv_reduce = fnv(fnv(fnv(a,b),c),d) result is fefffe6d , and Different Bits counts are 14 bits.
Rationale
In case of ethash algorithm, it can’t prevent ASIC forever.
And, current ethash algorithm’s FNV function is deprecated.
So, It needs to be upgraded and it will make current ethash based ASICs obsolete.
And current TETHASHV1 FNV1A implementation is based on most of ethash , which is verified for a long time.
Another propose of big differencing the Ethash algorithm need to crypto analysis for a long times and need to GPU code optimization times.
Verification and Optimization timeline Examples
original ethminer (2015) -> claymore optimized miner (2016) [1year]
genoil ethminer (2015) -> ethereum-mining/ethminer (2017) [2year]
Test Results::
Tethash miner has 2~3% of hashrate degrade on GPU, due to more core computation time.
Copyright
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.