mmap_timing.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import time\n",
    "\n",
    "from multiprocessing.dummy import Pool as ThreadPool"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1176378172132"
      ]
     },
     "execution_count": 21,
     "metadata": {
      "bento_obj_id": "140487237778800"
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = np.memmap('/tmp/0.faissindex_invlists')\n",
    "data.size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def benchmark_read(data, nq, sz, nt): \n",
    "    pool = ThreadPool(nt)\n",
    "    addresses = np.random.randint(data.size - sz, size=nq)\n",
    "    \n",
    "    def get_data(ad): \n",
    "        return data[ad:ad + sz].copy()\n",
    "    \n",
    "    t0 = time.time()\n",
    "    pool.map(get_data, addresses)        \n",
    "    \n",
    "    t1 = time.time()    \n",
    "    rate = sz * nq / (t1 - t0)\n",
    "    print(f'read {nq} blocks of size {sz} in {nt} threads: {t1 - t0:.3f} s ({rate / (1<<20):.2f} MiB / s)')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "read 50 blocks of size 480000 in 4 threads: 2.173 s (10.53 MiB / s)\n",
      "read 50 blocks of size 480000 in 10 threads: 1.705 s (13.42 MiB / s)\n",
      "read 50 blocks of size 480000 in 20 threads: 1.714 s (13.35 MiB / s)\n",
      "read 50 blocks of size 480000 in 40 threads: 1.638 s (13.97 MiB / s)\n"
     ]
    }
   ],
   "source": [
    "# impact of nb of read threads\n",
    "nq = 50\n",
    "sz = 16 * 30000\n",
    "\n",
    "for nt in 4, 10, 20, 40:\n",
    "    benchmark_read(data, nq, sz, nt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "read 500 blocks of size 1 in 20 threads: 19.289 s (0.00 MiB / s)\n"
     ]
    }
   ],
   "source": [
    "# reading a single byte\n",
    "nq = 500\n",
    "sz = 1\n",
    "nt = 20\n",
    "\n",
    "benchmark_read(data, nq, sz, nt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "read 1 blocks of size 200000000 in 20 threads: 5.230 s (36.47 MiB / s)\n",
      "read 10 blocks of size 20000000 in 20 threads: 3.004 s (63.49 MiB / s)\n",
      "read 100 blocks of size 2000000 in 20 threads: 8.036 s (23.73 MiB / s)\n",
      "read 1000 blocks of size 200000 in 20 threads: 37.519 s (5.08 MiB / s)\n"
     ]
    }
   ],
   "source": [
    "# reading the same amount of data with different fragmentations\n",
    "totsz = int(200e6)\n",
    "for nq in 1, 10, 100, 1000: \n",
    "    sz = totsz // nq\n",
    "    nt = 20\n",
    "    benchmark_read(data, nq, sz, nt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "bento_stylesheets": {
   "bento/extensions/flow/main.css": true,
   "bento/extensions/kernel_selector/main.css": true,
   "bento/extensions/kernel_ui/main.css": true,
   "bento/extensions/new_kernel/main.css": true,
   "bento/extensions/system_usage/main.css": true,
   "bento/extensions/theme/main.css": true
  },
  "kernelspec": {
   "display_name": "matthijsfaiss (local)",
   "language": "python",
   "name": "matthijsfaiss_local"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.5+"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}