Enhanced version of the highest priority measures

zhangtony239 commented 3 months ago

经过和gpt的对峙，我发现把CPU调度模式改成fifo，添加IO最高优先级，可以进一步压榨vps。

FIFO：先进先出，比RR轮询的实时模式更激进，这意味着hysteria会一次性完成响应再释放CPU，而不是和其他实时调度共享CPU资源。实际测试中，这个让我的浏览器加载Google首页快了半圈。

IO最高优先级：嗯...至少我的1panel把网络负载算进了IO，应该起了作用。

After a confrontation with gpt, I found that changing the CPU scheduling mode to fifo and increasing the highest IO priority can further squeeze vps.

FIFO: First in, first out, more aggressive than the real-time mode of RR polling, which means that hysteria will release the CPU only after running the response once, instead of sharing CPU resources with other real-time schedulers. In actual tests, this made my browser load the Google homepage half a circle faster.

IO highest priority: Well... at least my 1panel counts the network load into IO, which should have played a role.

Figure_1

zhangtony239 commented 3 months ago

由于改进后使得我vps响应“过于准时”，一秒的测速等待已然触发了我链路上的QoS，我这次测试的结果是改为了3秒一次测得的。

Because the improvement made my vps respond "too punctually", the one-second speed test wait had already triggered the QoS on my link. The result of my test this time was changed to a measurement every 3 seconds.

import time,requests
import seaborn as sns
import matplotlib.pyplot as plt

url='http://cp.cloudflare.com/' #10mb测试：'http://cachefly.cachefly.net/10mb.test'
port='2080' #HTTP代理端口

print('当前目标地址：'+url)

delays=[]
loss=0

headers = {
    'Cache-Control': 'no-cache',
    'http': 'http://127.0.0.1:'+port,
    'https': 'http://127.0.0.1:2080'+port,
}

def delaytest(url):
    start = time.time()
    try:
        requests.get(url,headers=headers,timeout=3)
    except:
        print('连接超时(>3000ms)')
        loss += 1
    end = time.time()
    delays.append((end - start)*1000)
    print(str((end - start)*1000)+'ms')

for i in range(10):
    try:
        delaytest(url)
        time.sleep(3)
    except:
        continue

plt.rcParams['font.sans-serif'] = ['SimHei'] # 指定默认字体
plt.rcParams['axes.unicode_minus'] = False # 解决保存图像是负号'-'显示为方块的问题

sns.boxplot(delays)
plt.title(url)
plt.xlabel('中位数：'+str(sorted(delays)[len(delays)//2])+'ms'+'，丢包率：'+str(loss*10)+'%')
plt.ylabel('延迟时间（毫秒）')
plt.show()

zhangtony239 commented 3 months ago

完善了IO优先级和CPU优先级，并添加了内存优先级和防杀死。

Improved IO priority and CPU priority, and added memory priority and anti-kill.

zhangtony239 commented 3 months ago

……或许太过激了

apernet / hysteria-website

Enhanced version of the highest priority measures #30