status-im
/
research
mirror of https://github.com/status-im/research.git
2
0
Fork 0
Code Issues Projects Releases Wiki Activity

Updated selfish validating sim for casper

This commit is contained in:
vub 2016-12-30 14:41:50 -05:00
parent 803be67d0e
commit 2d1dd0227b
2 changed files with 165 additions and 91 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

246
casper_sm/test.py
View File

@ -1,97 +1,171 @@
# The purpose of this script is to test selfish-mining-like strategies
# in the randao-based single-chain Casper.
import random
CHECK_DEPTH = 4
BLOCK_TIME = 1
SKIP_TIME = 1
FORK_PUNISHMENT_COEFF = 1.5
# Penalty for mining a dunkle
DUNKLE_PENALTY = 1
# Penalty to a main-chain block which has a dunkle as a sister
DUNKLE_SISTER_PENALTY = 0.8
attacker_share = 0.45
# Attacker stake power (out of 100). Try setting this value to any
# amount, even values above 50!
attacker_share = 40
# A simulated Casper randao
def randao_successor(parent, index):
return (((parent ^ 53) + index) ** 3) % (10**20 - 11)
# We categorize "scenarios" by seeing how far ahead we get a chain
# of 0-skips from each "path"; if the path itself isn't clear then
# return zero
heads_of_interest = ['', '1']
# Only scan down this far
scandepth = 4
# eg. (0, 2) means "going straight from the current randao, you
# can descend zero, but if you make a one-skip block, from there
# you get two 0-skips in a row"
scenarios = [None, (0, 1), (0, 2), (0, 3), (0, 4)]
# For each scenario, this is the corresponding "path" to go down
paths = ['', '10', '100', '100', '1000']
# Determine the scenario ID (zero is catch-all) from a chain
def extract_scenario(chain):
chain = chain.copy()
o = []
for h in heads_of_interest:
# Make sure that we can descend down "the path"
succeed = True
for step in h:
if not chain.can_i_extend(int(step)):
succeed = False
break
chain.extend_me(int(step))
if not succeed:
o.append(0)
else:
# See how far down we can go
i = 0
while chain.can_i_extend(0) and i < scandepth:
i += 1
chain.extend_me(0)
o.append(i)
if tuple(o) in scenarios:
return scenarios.index(tuple(o))
else:
return 0
# Class to represent simulated chains
class Chain():
def __init__(self, randao=0, time=0, length=0, me=0, them=0):
self.randao = randao
self.time = time
self.length = length
self.me = me
self.them = them
def copy(self):
return Chain(self.randao, self.time, self.length, self.me, self.them)
def can_i_extend(self, skips):
return randao_successor(self.randao, skips) % 100 < attacker_share
def can_they_extend(self, skips):
return randao_successor(self.randao, skips) % 100 >= attacker_share
def extend_me(self, skips):
new_randao = randao_successor(self.randao, skips)
assert new_randao % 100 < attacker_share
self.randao = new_randao
self.time += skips
self.length += 1
self.me += 1
def extend_them(self, skips):
new_randao = randao_successor(self.randao, skips)
assert new_randao % 100 >= attacker_share
self.randao = new_randao
self.time += skips
self.length += 1
self.them += 1
def add_my_dunkles(self, n):
self.me -= n * DUNKLE_PENALTY
self.them -= n * DUNKLE_SISTER_PENALTY
def add_their_dunkles(self, n):
self.them -= n * DUNKLE_PENALTY
self.me -= n * DUNKLE_SISTER_PENALTY
# randao_results[1101] = 1 -> if validator path is 1, 0, 1 then
# a colluding node will make the next block.
# randao_results[10011] = 0 -> if validator path is 1, 1, 0, 0
# then a non-colluding node will make the next block.
randao_results = [None, None]
for i in range(2, 2**CHECK_DEPTH):
randao_results.append(1 if random.random() < attacker_share else 0)
def update_randao():
for i in range(2, len(randao_results) / 2):
randao_results[i] = randao_results[i * 2]
for i in range(len(randao_results) / 2, len(randao_results)):
randao_results[i] = 1 if random.random() < attacker_share else 0
for strat_id in range(2**len(scenarios)):
# Strategy map: scenario to 0 = publish, 1 = selfish-validate
strategy = [0] + [((strat_id // 2**i) % 2) for i in range(1, len(scenarios))]
# 1 = once we go through the selfish-validating "path", reveal it instantly
# 0 = don't reveal until the "main chain" looks like it's close to catching up
insta_reveal = strat_id % 2
# Strategy map: number of blocks belonging to colluding node in
# the 0, 0, 0... chain -> 0 = publish, 1 = wait and compete, 2 = don't
# publish
print 'Testing strategy: %r, insta_reveal: %d', (strategy, insta_reveal)
for strat_id in range(2**CHECK_DEPTH):
strategy = [0]
k = strat_id
for _ in range(CHECK_DEPTH):
strategy.append(k % 2)
k /= 2
my_revenue = 0.
their_revenue = 0.
print 'Testing strategy:', strategy
pubchain = Chain(randao=random.randrange(10**20))
time = 0
while time < 200000:
child_0 = randao_results[2]
child_1 = randao_results[3]
situation = 0
q = 2
while situation < len(strategy) - 2 and randao_results[q] == 1:
situation += 1
q *= 2
if child_0 == 0:
their_revenue += 1
time += BLOCK_TIME
elif strategy[situation] == 0:
my_revenue += 1
time += BLOCK_TIME
elif strategy[situation] == 2:
their_revenue += 1
time += BLOCK_TIME + SKIP_TIME
else:
assert situation >= 1
my_score = situation
my_time = situation + 1
their_time = 0
their_score = 0
update_randao()
while their_time <= situation:
if randao_results[2] == 0:
wait_time = BLOCK_TIME
elif randao_results[3] == 0:
wait_time = BLOCK_TIME + SKIP_TIME
else:
wait_time = BLOCK_TIME + SKIP_TIME * 2
while random.random() < attacker_share:
wait_time += SKIP_TIME
their_score += 1
their_time += wait_time
update_randao()
time += max(my_time, their_time)
while my_score > their_score + 1:
if random.random() < attacker_share:
my_score += 1
if random.random() > attacker_share:
their_score += 1
time += BLOCK_TIME
if my_score > their_score or (my_score == their_score and random.random() < 0.5):
my_revenue += my_score - FORK_PUNISHMENT_COEFF
their_revenue -= their_score * 0.5
while time < 100000:
# You honestly get a block
if pubchain.can_i_extend(0):
pubchain.extend_me(0)
time += 1
continue
e = extract_scenario(pubchain)
if strategy[e] == 0:
# You honestly let them get a block
pubchain.extend_them(0)
time += 1
continue
# Build up the secret chain based on the detected path
# print 'Selfish mining along path %r' % paths[e]
old_me = pubchain.me
old_them = pubchain.them
old_time = time
secchain = pubchain.copy()
sectime = time
for skipz in paths[e]:
skips = int(skipz)
secchain.extend_me(skips)
sectime += skips + 1
# Public chain builds itself up in the meantime
pubwait = 0
while time < sectime:
if pubchain.can_they_extend(pubwait):
pubchain.extend_them(pubwait)
pubwait = 0
else:
their_revenue += their_score - FORK_PUNISHMENT_COEFF
my_revenue -= my_score * 0.5
update_randao()
if my_revenue * 1.0 / (my_revenue + their_revenue) > attacker_share - 0.01:
print 'My revenue:', my_revenue / time
print 'Their revenue:', their_revenue / time
print 'My share:', my_revenue / (my_revenue + their_revenue)
print 'Griefing factor:', (their_revenue / time / (1 - attacker_share) - 1) / (my_revenue / time / attacker_share - 1)
pubwait += 1
time += 1
secwait = 0
# If the two chains have equal length, or if the secret chain is more than 1 longer, they duel
while (secchain.length > pubchain.length + 1 or secchain.length == pubchain.length) and time < 100000 and not insta_reveal:
if pubchain.can_they_extend(pubwait):
pubchain.extend_them(pubwait)
pubwait = 0
else:
pubwait += 1
if secchain.can_i_extend(secwait):
secchain.extend_me(secwait)
secwait = 0
else:
secwait += 1
time += 1
# Secret chain is longer, takes over public chain, public chain goes in as dunkles
if secchain.length > pubchain.length:
pubchain_blocks = pubchain.them - old_them
assert pubchain.me == old_me
pubchain = secchain
pubchain.add_their_dunkles(pubchain_blocks)
# Public chain is longer, miner deletes secret chain so no dunkling
else:
pass
# print 'Score deltas: me %.2f them %.2f, time delta %d' % (pubchain.me - old_me, pubchain.them - old_them, time - old_time)
gf = (pubchain.them - 100000. * (100 - attacker_share) / 100) / (pubchain.me - 100000 * attacker_share / 100)
print 'My revenue: %d, their revenue: %d, griefing factor %.2f' % (pubchain.me, pubchain.them, gf)

10
zksnark/bn128_pairing.py
View File

@ -53,15 +53,15 @@ def miller_loop(Q, P):
if ate_loop_count & (2**i):
f = f * linefunc(R, Q, P)
R = add(R, Q)
assert R == multiply(Q, ate_loop_count)
# assert R == multiply(Q, ate_loop_count)
Q1 = (Q[0] ** field_modulus, Q[1] ** field_modulus)
assert is_on_curve(Q1, b12)
nQ2 = (Q[0] ** (field_modulus ** 2), -Q[1] ** (field_modulus ** 2))
assert is_on_curve(nQ2, b12)
# assert is_on_curve(Q1, b12)
nQ2 = (Q1[0] ** field_modulus, -Q1[1] ** field_modulus)
# assert is_on_curve(nQ2, b12)
f = f * linefunc(R, Q1, P)
R = add(R, Q1)
f = f * linefunc(R, nQ2, P)
R = add(R, nQ2)
# R = add(R, nQ2) This line is in many specifications but it technically does nothing
return f ** ((field_modulus ** 12 - 1) / curve_order)
# Pairing computation