Sorting Performance Comparison CUDF vs Dask vs Dask.map_partitions

sorting_comparision.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import dask_cudf\n",
    "import cudf\n",
    "import os\n",
    "import time\n",
    "import dask.dataframe as dd\n",
    "import dask.array as da"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from dask_cuda import LocalCUDACluster\n",
    "from dask.distributed import Client,wait"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "cluster = LocalCUDACluster()\n",
    "client = Client(cluster)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_random_data(n_rows=1_000,n_keys_index_1=50_000,n_keys_index_2=20_000,n_keys_index_3=20_000):\n",
    "    \n",
    "    df = dd.concat([\n",
    "        da.random.random(n_rows).to_dask_dataframe(columns='x'),\n",
    "        da.random.randint(0, n_keys_index_1, size=n_rows).to_dask_dataframe(columns='index_1'),\n",
    "        da.random.randint(0, n_keys_index_2, size=n_rows).to_dask_dataframe(columns='index_2'),\n",
    "        da.random.randint(0, n_keys_index_3, size=n_rows).to_dask_dataframe(columns='index_3'),\n",
    "        \n",
    "    ], axis=1).persist()\n",
    "    gdf = df.map_partitions(cudf.from_pandas)\n",
    "    return gdf.persist()\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Native Dask"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "time taken = 4.509349584579468 , len of df = 80,000,000\n"
     ]
    }
   ],
   "source": [
    "df = create_random_data(80_000_000)\n",
    "df = df.repartition(npartitions=1).persist()\n",
    "_ = wait(df)\n",
    "\n",
    "st = time.time()\n",
    "df = df.sort_values(by =['index_1','index_2', 'index_3']).persist()\n",
    "_ = wait(df)\n",
    "et = time.time()\n",
    "print(f\"time taken = {et-st} , len of df = {len(df):,}\")\n",
    "del df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Map Partitions "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "time taken = 1.7103350162506104 , len of df = 80,000,000\n"
     ]
    }
   ],
   "source": [
    "df = create_random_data(80_000_000)\n",
    "df = df.repartition(npartitions=1).persist()\n",
    "_ = wait(df)\n",
    "\n",
    "st = time.time()\n",
    "df = df.map_partitions(lambda df:df.sort_values(by =['index_1','index_2', 'index_3'])).persist()\n",
    "_ = wait(df)\n",
    "et = time.time()\n",
    "print(f\"time taken = {et-st} , len of df = {len(df):,}\")\n",
    "del df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Native Cudf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "time taken = 1.4036340713500977 , len of df = 80,000,000\n"
     ]
    }
   ],
   "source": [
    "df = create_random_data(80_000_000)\n",
    "df = df.compute()\n",
    "\n",
    "st = time.time()\n",
    "df = df.sort_values(by =['index_1','index_2', 'index_3'])\n",
    "_ = wait(df)\n",
    "et = time.time()\n",
    "print(f\"time taken = {et-st} , len of df = {len(df):,}\")\n",
    "del df"
   ]
  }
 ],
 "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}