notebook/e2e-rdma.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "b08fc02e",
   "metadata": {},
   "outputs": [],
   "source": [
    "import re\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "def parse_DeepLight_log(lib):\n",
    "    fname = f\"../models/DeepLight/100G-results/{lib}/log.txt\"\n",
    "    with open(fname, \"r\", encoding='UTF-8') as f:\n",
    "        lines = [line for line in f]\n",
    "    acc_epoch_time = []\n",
    "    for line in lines:\n",
    "        line = line.strip()\n",
    "        x = re.search(\".*Training.*time: (.*)s\", line)\n",
    "        if x is not None:\n",
    "            epoch_time = float(x.group(1))\n",
    "            acc_epoch_time.append(epoch_time)\n",
    "    return acc_epoch_time[1]  \n",
    "\n",
    "def parse_LSTM_log(lib):\n",
    "    fname = f\"../models/LSTM/100G-results/{lib}/log.txt\"\n",
    "    with open(fname, \"r\", encoding='UTF-8') as f:\n",
    "        lines = [line for line in f]\n",
    "    acc_epoch_time = []\n",
    "    next_epoch = 0\n",
    "    for line in lines:\n",
    "        line = line.strip()\n",
    "        x = re.search(\".*ms/batch (.*) \\| loss .*\", line)\n",
    "        if x is not None:\n",
    "            epoch_time = float(x.group(1))/1000\n",
    "            acc_epoch_time.append(epoch_time)\n",
    "    num = len(acc_epoch_time)\n",
    "    acc_epoch_time = np.array(acc_epoch_time[int(num/2):])\n",
    "    return np.average(acc_epoch_time)\n",
    "\n",
    "def parse_NCF_log(lib):\n",
    "    fname = f\"../models/NCF/100G-results/{lib}/log.txt\"\n",
    "    with open(fname, \"r\", encoding='UTF-8') as f:\n",
    "        lines = [line for line in f]\n",
    "    acc_epoch_time = []\n",
    "    for line in lines:\n",
    "        line = line.strip()\n",
    "        x = re.search(\".*train time: (.*)\", line)\n",
    "        if x is not None:\n",
    "            epoch_time = float(x.group(1))\n",
    "            acc_epoch_time.append(epoch_time)\n",
    "    num = len(acc_epoch_time)\n",
    "    acc_epoch_time = np.array(acc_epoch_time[int(num/2):])\n",
    "    return np.average(acc_epoch_time)\n",
    "\n",
    "def parse_BERT_log(lib):\n",
    "    fname = f\"../models/BERT/100G-results/{lib}/log.txt\"\n",
    "    with open(fname, \"r\", encoding='UTF-8') as f:\n",
    "        lines = [line for line in f]\n",
    "    acc_tput = []\n",
    "    acc_epoch_time = []\n",
    "    for line in lines:\n",
    "        line = line.strip()\n",
    "        x = re.search(\".*step_loss .*step_time_ns : (.*)\", line)\n",
    "        if x is not None:\n",
    "            epoch_time = float(x.group(1))/1000000000\n",
    "            acc_epoch_time.append(epoch_time)\n",
    "    num = len(acc_epoch_time)\n",
    "    acc_epoch_time = np.array(acc_epoch_time[int(num/2):])\n",
    "    return np.average(acc_epoch_time)\n",
    "\n",
    "def parse_ResNet152_log(lib):\n",
    "    fname = f\"../models/CNN/100G-results/{lib}/ResNet152_log.txt\"\n",
    "    with open(fname, \"r\", encoding='UTF-8') as f:\n",
    "        lines = [line for line in f]\n",
    "    acc_time = []\n",
    "    i = 0\n",
    "    model=\"\"\n",
    "    for line in lines:\n",
    "        line = line.strip()\n",
    "        if i==0:\n",
    "            model = line\n",
    "            i = i + 1\n",
    "        else:\n",
    "            x = re.search(\"Epoch:\\s+\\[0\\]\\[([0-9]+)/([0-9]+)\\]\\s+Time\\s+([0-9]+\\.[0-9]+)\\s+\\(.*\\)\\s\", line)\n",
    "            if x is not None:\n",
    "                iter_no = int(x.group(1))\n",
    "                batch_time = float(x.group(3))\n",
    "                acc_time.append(batch_time)\n",
    "    num = len(acc_time)\n",
    "    acc_time = np.array(acc_time[int(num/2):])\n",
    "    return np.average(acc_time)\n",
    "\n",
    "def parse_VGG19_log(lib):\n",
    "    fname = f\"../models/CNN/100G-results/{lib}/VGG19_log.txt\"\n",
    "    with open(fname, \"r\", encoding='UTF-8') as f:\n",
    "        lines = [line for line in f]\n",
    "    acc_time = []\n",
    "    i = 0\n",
    "    model=\"\"\n",
    "    for line in lines:\n",
    "        line = line.strip()\n",
    "        if i==0:\n",
    "            model = line\n",
    "            i = i + 1\n",
    "        else:\n",
    "            x = re.search(\"Epoch:\\s+\\[0\\]\\[([0-9]+)/([0-9]+)\\]\\s+Time\\s+([0-9]+\\.[0-9]+)\\s+\\(.*\\)\\s\", line)\n",
    "            if x is not None:\n",
    "                iter_no = int(x.group(1))\n",
    "                batch_time = float(x.group(3))\n",
    "                acc_time.append(batch_time)\n",
    "    num = len(acc_time)\n",
    "    acc_time = np.array(acc_time[int(num/2):])\n",
    "    return np.average(acc_time)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "6d0bfe16",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x = ['DeepLight', 'LSTM', 'NCF', 'BERT', 'ResNet152', 'VGG19']\n",
    "speedups = []\n",
    "for l in x:\n",
    "    func = 'parse_'+l+'_log'\n",
    "    speedups.append(eval(func)('NCCL')/eval(func)('omnireduce'))\n",
    "fig, (ax) = plt.subplots(1,1)\n",
    "fig.set_size_inches(12,6)\n",
    "ax.bar(x, speedups, width=0.3, edgecolor='white', label='OmniReduce', color='#ff7f0e')\n",
    "for u, v in enumerate(speedups):\n",
    "    ax.text(u-0.1, v + 0.01, str(round(v,1)), fontsize=15, zorder=99)\n",
    "ax.tick_params(axis='both', which='major', labelsize=20)\n",
    "ax.axhline(y=1, linewidth=2, color='r', linestyle='--', label='NCCL')\n",
    "ax.set_ylabel('Speedup', fontsize=20)\n",
    "plt.legend(fontsize=20)\n",
    "plt.grid()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c8b40693",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}