course/hse-2026
3
Fork 1
Code Issues Pull Requests Actions Packages Activity

Add week 8 materials

Browse Source
This commit is contained in:
osukhoroslov committed 2026-11-04 19:23:39 +03:00
1 parent a987b82834
commit 0bf71ed374
38 files changed
+2908
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

materials/08-scaling/seminar/load-balancing-lab/Lab.ipynb
+569
View File
@@ -0,0 +1,569 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "9ba1cf58",
"metadata": {},
"source": [
"# Балансировщики нагрузки: лабораторная работа"
]
},
{
"cell_type": "markdown",
"id": "0d8f9518",
"metadata": {},
"source": [
"В этой лабораторной работе сравним алгоритмы балансировки по задержкам запросов. Будем менять нагрузку, число балансировщиков и актуальность сведений о серверах. Перед запуском выполните [инструкции подготовки](readme.md): Docker-образ нужно собрать из текущих исходников."
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "7c48777d",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<IPython.core.display.Image object>"
]
},
"execution_count": 1,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"from IPython.display import Image\n",
"Image(filename='media/balancers.png') "
]
},
{
"cell_type": "markdown",
"id": "9cfcf69f",
"metadata": {},
"source": [
"Все балансировщики распределяют запросы по **одному общему набору серверов**. Это дискретно-событийная симуляция на Rust с библиотекой simcore: серверы и балансировщики моделируются обработчиками событий внутри одного процесса.\n",
"\n",
"- Запрос имеет числовой ID отправителя и длительность обслуживания. На медленном сервере длительность **умножается** на коэффициент замедления.\n",
"- Каждый сервер обслуживает запросы последовательно, остальные ждут в очереди.\n",
"- Входящий запрос попадает на случайный балансировщик. Состав серверов в течение эксперимента не меняется.\n",
"- Раз в `sync_interval` секунд балансировщик обновляет снимок числа незавершённых запросов на каждом сервере, включая обслуживаемый запрос. Между обновлениями снимок может устареть.\n",
"\n",
"`sync_interval` — интервал обновления сведений, а не время обработки запроса балансировщиком. Затраты на выбор сервера, сетевые задержки и стоимость проверок в этой модели не учитываются.\n",
"\n",
"На всех графиках по оси X отложено начало окна `t`, по оси Y — **p99 времени ответа** для запросов, пришедших в интервал `[t, t + 60)` секунд. Время ответа включает ожидание и обслуживание. Чтобы получить p99, длительности сортируют и берут элемент с номером `ceil(0.99 × n)` при нумерации с единицы. Например, для 100 запросов это 99-е значение, а не максимум. Ось Y логарифмическая.\n",
"\n",
"Симуляция создаёт запросы до конца последнего окна, затем дожидается завершения всех очередей. Поэтому даже долгие ответы учитываются полностью. Параметр `--time-limit` задаёт последнюю отметку на оси X, а не момент принудительного завершения запросов."
]
},
{
"cell_type": "markdown",
"id": "79e49154-fa8e-4bad-9377-b72b5350deb4",
"metadata": {},
"source": [
"## Вспомогательный код"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "ebd17f6c",
"metadata": {},
"outputs": [],
"source": [
"import json\n",
"import matplotlib.pyplot as plt\n",
"import numpy as np\n",
"import subprocess\n",
"\n",
"from pathlib import Path\n",
"\n",
"# Сначала соберите образ по инструкции в readme.md.\n",
"# Для локальной сборки и запуска через Cargo задайте False.\n",
"USE_DOCKER = True\n",
"DOCKER_IMAGE = 'distsys-load-balancing-lab:local'\n",
"\n",
"\n",
"def run_experiment(name, n_balancers=1, settings=None, balancers=None,\n",
" sync_interval=0.01, seed=123):\n",
" settings = list(settings or [])\n",
" if balancers is None:\n",
" balancers = ['Hash', 'Random', 'RoundRobin', 'LeastLoaded']\n",
" config_path = Path('.tmp-config.json')\n",
" result_path = Path('.tmp-result.txt')\n",
" configs = [{'balancer': balancer, 'n_balancers': n_balancers,\n",
" 'sync_interval': sync_interval} for balancer in balancers]\n",
" config_path.write_text(json.dumps(configs), encoding='utf-8')\n",
" args = ['--servers', '100', '--seed', str(seed), '--dump', result_path.name,\n",
" '--config', config_path.name, *settings]\n",
" try:\n",
" if USE_DOCKER:\n",
" command = ['docker', 'run', '--rm', '--mount',\n",
" f'type=bind,source={Path.cwd()},target=/lab',\n",
" DOCKER_IMAGE, *args]\n",
" else:\n",
" command = ['cargo', 'run', '--locked', '--release', '--', *args]\n",
" subprocess.run(command, check=True)\n",
" results = result_path.read_text(encoding='utf-8').splitlines()\n",
" finally:\n",
" config_path.unlink(missing_ok=True)\n",
" result_path.unlink(missing_ok=True)\n",
"\n",
" plt.figure(figsize=(12, 5))\n",
" plt.title(name)\n",
" plt.yscale('log')\n",
" plt.ylabel('p99 времени ответа, с (окно 60 с)')\n",
" plt.xlabel('Начало окна t, с')\n",
" for i in range(0, len(results), 3):\n",
" times = np.fromstring(results[i + 1], sep=',')\n",
" response_times = np.fromstring(results[i + 2], sep=',')\n",
" plt.plot(times, response_times, label=results[i])\n",
" plt.legend(loc='upper left')\n",
" plt.tight_layout()\n",
" plt.show()\n"
]
},
{
"cell_type": "markdown",
"id": "7bbba437",
"metadata": {},
"source": [
"## Эксперимент 1: централизованный балансировщик"
]
},
{
"cell_type": "markdown",
"id": "fcb15af2",
"metadata": {},
"source": [
"Начнём с четырёх алгоритмов:\n",
"\n",
"- **Hash** закрепляет отправителя за сервером. В симуляции это `sender_id % num_servers` — упрощённая модель детерминированного выбора по клиенту.\n",
"- **Random** каждый раз выбирает случайный сервер.\n",
"- **RoundRobin** перебирает серверы по кругу.\n",
"- **LeastLoaded** выбирает сервер с наименьшим числом незавершённых запросов по последнему снимку. Это не Least Connections: TCP-соединения здесь не моделируются.\n",
"\n",
"Сначала работает один балансировщик с обновлением снимков раз в **10 мс**. Сравниваем p99 времени ответа, определённый выше. Меньшее значение означает меньшую задержку на границе самых медленных 1% запросов."
]
},
{
"cell_type": "markdown",
"id": "bb9119ac",
"metadata": {},
"source": [
"Запустим эксперимент на 100 одинаковых серверах и 200 отправителях с сопоставимой интенсивностью запросов. При одинаковых параметрах нагрузки и `seed` все запуски используют одинаковый входной поток; генераторы выбора балансировщика и бэкенда отделены от генератора нагрузки."
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "5cc16a68",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('balanced workload')"
]
},
{
"cell_type": "markdown",
"id": "475a053a",
"metadata": {},
"source": [
"Уменьшим число отправителей до 10, сохранив 100 серверов. Сколько серверов сможет использовать Hash при таком выборе? Нужны ли коллизии, чтобы нагрузка оказалась неравномерной?"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "738b357c",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('balanced workload (10 senders)', settings=['--senders', '10'])"
]
},
{
"cell_type": "markdown",
"id": "d1b44c0b",
"metadata": {},
"source": [
"Теперь вернём 200 отправителей, но один из них будет посылать запросы в 50 раз чаще остальных. Проследите, куда попадёт его нагрузка при разных алгоритмах."
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "ab3d360c",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('skewed workload', settings=['--preset', 'skewed'])"
]
},
{
"cell_type": "markdown",
"id": "34925ed0",
"metadata": {},
"source": [
"- Почему закрепление отправителя за сервером ухудшает задержки в этих двух случаях?\n",
"- Когда такая привязка полезна, например для локального кеша или состояния сессии?\n",
"- Чем равномерное распределение числа отправителей отличается от равномерного распределения нагрузки?"
]
},
{
"cell_type": "markdown",
"id": "1bccea52",
"metadata": {},
"source": [
"Пока все серверы были одинаковыми. Вернём равномерный поток от 200 отправителей и замедлим 5% серверов в пять раз."
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "0372fc52",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "iVBORw0KGgoAAAANSUhEUgAABaAAAAJYCAYAAACUxkEyAAAAOnRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjEwLjcsIGh0dHBzOi8vbWF0cGxvdGxpYi5vcmcvTLEjVAAAAAlwSFlzAAASdAAAEnQB3mYfeAABAABJREFUeJzs3Qd4FOXaBuBnZ7an95AAofemUkRQEEERO9iOFRv+9t4LYMeKysEjFvAcu9iwoKiAoiBFQKT3GiC9Z/v81/vFjQlJIMAGUp77uvbsZmZ25pvZTQ4+++77mQzDMEBEREREREREREREFGJaqHdIRERERERERERERMQAmoiIiIiIiIiIiIjqDCugiYiIiIiIiIiIiKhOMIAmIiIiIiIiIiIiojrBAJqIiIiIiIiIiIiI6gQDaCIiIiIiIiIiIiKqEwygiYiIiIiIiIiIiKhOMIAmIiIiIiIiIiIiojrBAJqIiIiIiIiIiIiI6gQDaCIiIiIiIiIiIiKqEwygiYiIiIiIiIiIiKhOMIAmIiIiIiIiIiIiojrBAJqIiIiIiIiIiIiI6gQDaCIiIiIiIiIiIiKqEwygiYiIiIiIiIiIiKhOMIAmIiIiIiIiIiIiojrBAJqIiIiIiIiIiIiI6gQDaCIiIiIiIiIiIiKqEwygiYiIiIiIiIiIiKhOMIAmIiKiRmHr1q0wmUwYPXp0nR5HjjF48GA0JXK+ct4NWatWrdStNubOnavOd9y4cWio51DR2LFjYbfbsWPHjsM61rRp09R1kft9zZo1CyeccAKio6PVNueee275uiVLlmDYsGGIj49X63r16nXQ50BNQ0lJCZKTk3HZZZcd7aEQERFRCJlDuTMiIiIiIqo/JHR+7rnnMGbMGLRo0aLOPvw555xzVPh89dVXIzIyEp06dVLrCgoKcMYZZ8DlcuHyyy9XIbQEjA2VfDhx8sknq1C/vn1A0Rg4nU488MADuOOOO3DbbbehT58+R3tIREREFAIMoImIiIiIGqnHH38cbrcb995772Hv67zzzsPxxx+PZs2aVVr+448/qoD5hRdewCWXXFJp3aJFi5CRkYEnn3wSDz744GGPgRq/66+/HuPHj8dDDz2kKuuJiIio4WMLDiIiIiKiRig/Px/vvfceTjnlFDRv3vyw9xcVFaUqm+W+ovT0dHWfkpJS5Tn7W0dUHWkXc9FFF6kPNjZs2MCLRERE1AgwgCYiIqJGZ+3ataoHbWxsLMLCwjBw4MBqK+kkoJP2BEOGDFEBndVqRUJCAs4++2wsWLCg1seTkO2xxx7DgAEDVHsB2Y8EblINunr16v32q5bHF198sWpNIMFL79698fXXX9d4rI8++kgFinJusr306f3Xv/6l+uzu64MPPlDtAqQ1gmzbuXNnPPHEE6oitjoffvghjjvuODgcDiQmJqqWCcEAsbb69++vzr+4uLjS8kGDBqlzvuaaayotX7NmjVp+xRVXVHlt5Kv4HTt2VGOPiYnBaaedpkKp/fVslopbafkg10eWyfXdn/fffx82m01dmwNtKyQQk7GmpqaWv87yc3VB2cG+L4RhGJg0aRK6du2qzluOc/PNN6vrcbDk9ZeeuhLmheJY+/aADl53aUch5L0mPwe3kfsrr7xSrbvqqqsqrQuS8T399NOqL7T8roaHh6v3kIz9UF/ng3nfB3u6Z2VlqTYlUt0t7we5JlOnTq20rfy+yn6FVOgGz0duMrYDmTdvHs466yz1t0aOIe8JqSiXfe0rVNdF/hbKdZDfZ5/PV+24brjhBrXtvn935LlyztK6Rd67SUlJ6r27bt26KvuQ7WQfmzdvxquvvooePXqovyPBfvnyXnvnnXdUn3D5Gyuvi+xXfqflb9q+5G+iPOftt98+4HUlIiKi+o8tOIiIiKhR2bJliwpqunfvrr7KvXv3bhVwnH766SpsrBjGSfgpX/M+6aSTVGgjIef27dsxY8YMzJw5E1999RWGDx9+wGP+8ssveOaZZ1Q4NWrUKBUWSSA5ffp0ta/ffvsNPXv2rPK8bdu2oW/fvmjTpo0Ke3NyctRYpZ+uBK3BsEtIGCMhnoQ4ElaPHDlSBTk7d+7EnDlzVFAr4XWQ9OKVAE3CLhmThFC///47HnnkEfz000/44YcfYDb/80/Bl156CXfeeafaTgJVuf/+++9VYLRvxev+SDgux5GwLXjtJEyTZUKOXVHwZ3leUF5engptJaSVHrC33367Cgg//vhjnHrqqXjttdfUa7sv+dBAQjv5wEHOX54jwVlNnn32Wdx///3qHOV1ktBufxYvXoyhQ4eisLBQfUjRpUsXFdK9++67+PLLL9VrVrFn7aG8L+RcX3nlFRWESiBqsVjUvhcuXAiPx7Pf89lXMKyX61Gdwz2WfPgh4bMEoD///LMKm4MTF0pwKuuWL1+u9inv6eDkg8F7eZ3lw59ly5bh2GOPVa9ZIBBQ7zsJOletWqWC44N5nQ/2fV/x/Sb7OP/881VQ/cknn6h9aZpWHqIHJ1aU30H5QKXiZKQHmhzyu+++U39jpD+2vHck7Jffd/kbNHny5PIQP9TXRY4nf/OmTJmi/qZJAF6RnKv8zZFwueLfOhmv/I3xer3qOe3atVN/az777DN888036m+OjG1f0rdZfvflXEeMGAFd19Vy+TsrY2vdujUuvPBC9TdF/jbL75Rc630/JJG/i/J+lNdLnkdEREQNnEFERETUCGzZssWQf9rI7e677660bvHixYbZbDaio6ON/Pz88uV5eXlGZmZmlX3t2LHDaNasmdGpU6cq62T/gwYNqrRs7969RkFBQZVtly9fboSFhRnDhw+vcazjxo2rtO67775Ty08//fRKy19//XW1vE+fPmrcFfl8PiM9Pb3856lTp6ptzzvvPKOkpKTStmPHjlXrJk6cWGk8FovFiImJUY+D/H6/MXLkyPKx1sZPP/1U5TUIntOwYcPU/caNG8vXnXvuuWrZ9u3by5eNGTNGLZP7QCBQvnz9+vVGZGSkYbVaK41zzpw55WP8z3/+U+240tLS1C14XjfffLPaXs6vtLS00rbB/cm1CpJxyPtBlr/77ruVtv/www/V8o4dO6p9H+r74rffflP7adu2rZGdnV2+XMZ3/PHHq3XBc6iNpKQkdb0qXsPDOVbwfSX31b2n5Lrtq6bniCuvvFKtmzBhQqXlMobTTjvNMJlMxrJly2r9Oh/s+14E93fNNdeo36OgVatWGbquG507dz7ge6M2gr9H8trva9+/QaG+LvPnz1frRo0aVWXdxx9/rNbdeeed5ctycnLU38q4uDh1HSr666+/1Hv3mGOOqXbMKSkpxubNm6scJzY21khNTTWKi4sPeP5BvXr1MjRNq/Z3iIiIiBoWBtBERETUKARD3aioqGoDi2BAMm3atFrt75ZbblHbb9u27YAB9P6cddZZhs1mMzweT5WxSsBXMfQKatmypQp/KurWrZt6ztKlSw94TAluJHDPzc2tsk6OJ/uWIDvoiSeeUPt+9NFHq2y/adMmFQLVNoCWkMxut1cKqO655x41nmAQJmG6kLBWgq727duXb+t2uw2n02mEh4dXCkaDHn74YbWP8ePHVwng5LxrEgygZXwSUMr28hpXDIz3FzL++uuvaln//v2r3f/AgQPV+p9//vmQ3xfXXnut2sfbb79d45hqG0DLdZTtK17big7lWKEMoLOyslTA27t372rHJ0GtPE/eO7V9nQ/2fS9kf/J+q/jBVNBJJ52k1hcWFoYsgF63bt1+t6uL6yI6dOigPrjZ93fqjDPOUM/9888/y5dJSC/LJk2aVO2+br/9drW+Yjgd/Pu6b8BfMYBu1aqV4XK5jNqSD2hkn2vWrKn1c4iIiKh+YgsOIiIialTka+ERERFVlsvX5eWr8/K19uBX6oW0QXj55ZfV19czMjJU64GKdu3ahZYtWx7wuPK19P/85z+qF7N89X3ffquyTFodVCStCIJfUa9IeqNW7EEt/ZRXrlypviZ/zDHH7Hcc0u7izz//VG06Jk6cWO020n9WvvoftHTpUnUvbQX2Je1BZDzSLqQ2pLertLSQr+hnZ2cjLi4Os2fPVq0ppDWKnIO0QpCWD3JcaTdQ8ev30l9WzkFaIlTXEkNaE0j7AXkd9yVf29+f0tJS1epDru2ECRNw7733oraC10iOXx1Z/uuvv6pxSUuXQ3lf7O91kLYK1b1XaiLXXkhbmf2dTyiOdSik9YLf7y/vXbwvaf0gKr5P9/c6H8r7Pqh9+/aqVcW+5H0vcnNzVfuUw3HppZeq9hX9+vVT73dpyyLv8X0nhwz1dQmSv3nSBkP6vN94441q2d69e1VbD/mbIj2bg4J/e+R6VjeG9evXl49B2tDUZgxy/tIbWraXFhzyvpO/B/tr7xP8/ZffESIiImrYGEATERFRoyIBZ3Vkwi9RcYK1zz//XPV8ldB02LBhaNu2rZrwS/q+Bvva1jRhX0USYEs/XQn7ZD8SWDudThUiffHFFyrIqW4/0p+2OtKjVnq+BklIK6Rv7IFIWCaFnZmZmdVOblad4DXZ37WrbQAtJOSV0FlCaHksoeyDDz5YHtRKb2IZY3X9n4Nj2TesDwouD16Tfce5P9K7WYJXCRtl8rODcSjjOtj3xf5eB3lPSLhaWzIBnHC5XPs9n1Ac61AEA3IJXOVWk6Kiolq9zofyvq/N76GQQPhwST9lmeTvhRdeUBPrvf7662q5TPopPY7l/VEX1yVI+rpLH2z5EC4YQL/33nvqA5GKH8hVHMMbb7yx33M6mDFIj3n5MEv6c0tfdLnJ9ZU+0XJNpMd0dR8YVXwvExERUcPFAJqIiIgaFanqq86ePXvUfcWKOwlkZOIxqU7t3Llzpe1lkjsJoA9EAhypEpTgRcLNfQPKipXMhyoYkEk19oEEz0+qGoNVrrV9jly7rl271njtaitYJSxBs4T5EqYHQ2ZZ98EHH6jwVQJoCWMrTrYYHEtLine truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('balanced workload (different servers)', settings=['--different-servers'])"
]
},
{
"cell_type": "markdown",
"id": "9c8afb20",
"metadata": {},
"source": [
"- Почему алгоритмы, не учитывающие очереди, накапливают запросы на медленных серверах?\n",
"- Насколько LeastLoaded помогает в этой ситуации и как на результат влияет период обновления снимка?"
]
},
{
"cell_type": "markdown",
"id": "ce26e221",
"metadata": {},
"source": [
"## Эксперимент 2: независимые балансировщики"
]
},
{
"cell_type": "markdown",
"id": "ed6ad1a3",
"metadata": {},
"source": [
"Теперь несколько балансировщиков используют общий набор серверов. У каждого свой снимок состояния, который обновляется раз в **5 секунд**. В этой модели обновления происходят одновременно, поэтому все балансировщики видят одинаковые снимки. Запросы по-прежнему распределяются между балансировщиками случайно.\n",
"\n",
"По умолчанию у экземпляров Random и PowerOfK разные воспроизводимые последовательности случайных чисел, а у RoundRobin — случайные начальные позиции. Параметр `seed` позволяет воспроизвести эксперимент.\n",
"\n",
"По сравнению с первым экспериментом меняются и число балансировщиков, и период обновления. Сначала посмотрим на один балансировщик с периодом 5 секунд, чтобы отделить эти эффекты."
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "623b10ba",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('balanced workload (1 balancer, updates every 5 s)', sync_interval=5)"
]
},
{
"cell_type": "markdown",
"id": "b2578704",
"metadata": {},
"source": [
"- Зачем распределять входящий поток между несколькими балансировщиками?\n",
"- Что изменилось при переходе от обновлений раз в 10 мс к обновлениям раз в 5 секунд при одном балансировщике?"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "a5f41eba",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('balanced workload (5 balancers)', n_balancers=5, sync_interval=5)"
]
},
{
"cell_type": "markdown",
"id": "59fb2833",
"metadata": {},
"source": [
"- Почему LeastLoaded направляет много запросов на один сервер между обновлениями? Почему при одинаковых снимках и входном потоке его кривая не меняется с числом балансировщиков?\n",
"- Как изменяется p99 у Random и RoundRobin? Не ожидайте одинакового порядка кривых при любых параметрах нагрузки.\n"
]
},
{
"cell_type": "markdown",
"id": "17865058",
"metadata": {},
"source": [
"## Эксперимент 3: Power of Two Choices и Power of k Choices"
]
},
{
"cell_type": "markdown",
"id": "158f1f53",
"metadata": {},
"source": [
"С устаревшими снимками глобальный минимум может привлекать слишком много запросов от разных балансировщиков. Попробуем выбирать минимум только среди **двух случайных серверов** — Power of Two Choices. Более общий вариант с k кандидатами называется Power of k Choices.\n",
"\n",
"Это не гарантия улучшения для любой нагрузки: результат зависит от актуальности снимков, длительности запросов и различий между серверами. Проверим алгоритм на том же наборе из пяти балансировщиков. Подробнее о методе: [The Power of Two Choices in Randomized Load Balancing](http://www.eecs.harvard.edu/~michaelm/postscripts/handbook2001.pdf)."
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "98374a5a",
"metadata": {},
"outputs": [],
"source": [
"exp3_balancers = ['Hash', 'Random', 'RoundRobin', 'LeastLoaded', 'PowerOfK[k=2]']"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "64609ded",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('power of two choices', n_balancers=5, sync_interval=5, balancers=exp3_balancers)"
]
},
{
"cell_type": "markdown",
"id": "44c62e12",
"metadata": {},
"source": [
"Увеличим число кандидатов k. Больший выбор повышает шанс найти свободный сервер, но при устаревших снимках может усилить концентрацию запросов на одних и тех же кандидатах. Какой эффект виден на этих графиках? Повторите эксперимент с другим периодом обновления или seed."
]
},
{
"cell_type": "code",
"execution_count": 14,
"id": "5f1c5da8",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('power of K choices', n_balancers=5, sync_interval=5, balancers=['PowerOfK[k=2]', 'PowerOfK[k=3]', 'PowerOfK[k=5]', 'PowerOfK[k=10]'])"
]
},
{
"cell_type": "markdown",
"id": "198d4745",
"metadata": {},
"source": [
"PowerOfK выбирает по периодически обновляемому снимку. Вариант **PowerOfKWithUpdates** сравнивает актуальные очереди только у выбранных k серверов перед каждым решением. В симуляции это прямое чтение состояния: стоимость и задержка таких проверок не моделируются. Сравните варианты и обсудите, какой дополнительный обмен потребовался бы в реальной системе."
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "ecb9a473",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('do updates matter?', n_balancers=5, sync_interval=5, balancers=['PowerOfK[k=2]', 'PowerOfKWithUpdates[k=2]'])"
]
},
{
"cell_type": "markdown",
"id": "cffbb495",
"metadata": {},
"source": [
"Повторим сравнение при наличии медленных серверов. Сохраняется ли преимущество алгоритмов, учитывающих очередь? Как на результаты влияют выбранный seed и частота обновления состояния?"
]
},
{
"cell_type": "code",
"execution_count": 16,
"id": "ddcbc891",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('power of two choices', n_balancers=5, sync_interval=5, balancers=exp3_balancers, settings=['--different-servers'])"
]
},
{
"cell_type": "markdown",
"id": "correlated-selectors-intro",
"metadata": {},
"source": [
"## Дополнительно: коррелированные выборы балансировщиков\n",
"\n",
"Этот эксперимент можно выполнить после основной части. Сравните его с запуском `balanced workload (5 balancers)` из эксперимента 2: число балансировщиков, поток запросов и период обновления снимков остаются прежними.\n",
"\n",
"Флаг `--correlated-selectors` задаёт всем балансировщикам одинаковый seed выбора сервера. Поэтому у экземпляров Random совпадают последовательности случайных чисел, а у RoundRobin — начальные позиции. Это отличается от общего параметра `seed`: без флага он воспроизводит эксперимент, но экземпляры получают разные последовательности выбора.\n"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "5e3a734f",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 1200x500 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"run_experiment('correlated selectors (5 balancers)', n_balancers=5, sync_interval=5,\n",
" settings=['--correlated-selectors'])"
]
},
{
"cell_type": "markdown",
"id": "correlated-selectors-questions",
"metadata": {},
"source": [
"- Как изменились p99 у Random и RoundRobin по сравнению с независимой инициализацией? Почему одинаковые последовательности могут создавать перекосы нагрузки?\n",
"- Почему результаты Hash и LeastLoaded не изменились? Используют ли они генератор случайных чисел при выборе сервера?\n",
"\n",
"В этом прогоне особенно заметен рост p99 у Random. Запросы поступают на балансировщики случайно, поэтому их последовательности продвигаются с разной скоростью: одинаковый seed не означает, что все балансировщики всегда выбирают один сервер одновременно.\n"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.13.7"
}
},
"nbformat": 4,
"nbformat_minor": 5
}