Last active
June 22, 2021 00:08
-
-
Save alecco/8a1f0f56eeae5b344a9f299164ff600c to your computer and use it in GitHub Desktop.
Raft dueling candidates with power distribution
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# What if n nodes are assigned a random tick wait (of 1 to 10) | |
# to wait to become candidates when the previous leader is down? | |
# What is the chance of 2 candidates having the same random | |
# wait (birthday paradox) AND this wait to be the shortest wait. | |
# What's the amount of nodes with ~XX% chance of dueling candidates? | |
# Based on: | |
# https://codecalamity.com/the-birthday-paradox-the-proof-is-in-the-python/ | |
# | |
# For power distribution | |
# | |
# For [1,10] | |
# Shape 50% | |
# 1 13 (uniform) | |
# 2 64 | |
# 3 138 | |
# 4 235 | |
# 5 295 | |
# | |
from random import randint | |
import matplotlib.pyplot as plt | |
from multiprocessing import Pool, cpu_count | |
from bisect import bisect | |
import matplotlib.pyplot as plt | |
import numpy as np | |
WAIT_MIN = 1 | |
WAIT_MAX = 10 | |
MAX_NODES = 250 # Make it close to 100% chance | |
PERC_THRESHOLD = 50 # Find 50% chance of dueling | |
SHAPE_PARAM = 4 # Power distribution shape | |
rng = np.random.default_rng(12345) | |
def random_timeouts(nodes): | |
return (np.random.power(SHAPE_PARAM, nodes) * WAIT_MAX + 1).astype(int) | |
def determine_probability(nodes, run_amount=10000): | |
duplicate_minimum_found_cnt = 0 | |
for _ in range(run_amount): | |
timeouts = random_timeouts(nodes) | |
# minimum value is repeated | |
try: | |
duplicate_minimum_found_cnt += (np.count_nonzero(timeouts == timeouts.min()) > 1) | |
except ValueError: | |
# empty | |
pass | |
# return node number for sorting all subprocess results | |
return nodes, duplicate_minimum_found_cnt/run_amount * 100 | |
def plot_candidate_probabilities(max_nodes, perc_threshold=PERC_THRESHOLD): | |
with Pool(processes=cpu_count()) as p: | |
percent_chances = p.map(determine_probability, range(max_nodes)) | |
res = sorted(percent_chances, key=lambda x: x[0]) # results by x | |
res_y = [z[1] for z in res] # y values (discard x, now implicit) | |
plt.plot(res_y) # plot it | |
plt.xlabel("Number of nodes") | |
plt.ylabel('Chance of dueling candidates (Percentage)') | |
idx_threshold = bisect(res_y, perc_threshold) | |
if max_nodes > idx_threshold: | |
print(f"{perc_threshold}% cutoff is {idx_threshold}: {res_y[idx_threshold]}%") | |
plt.axvline(x=idx_threshold, color='red') | |
plt.text(idx_threshold, 0, str(idx_threshold), color='red') | |
else: | |
print("all below 90%, try again") | |
plt.savefig(f"dueling_candidates_power_{SHAPE_PARAM}.png") | |
if __name__ == '__main__': | |
plot_candidate_probabilities(MAX_NODES) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Shape 1 (uniform):
Shape 2:
Shape 3:
Shape 4:
Shape 5: