Experiments for caching cudf DataFrame objects that will be re-read many times

CPU_caching_experiment.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import cudf\n",
    "import pickle\n",
    "from io import BytesIO\n",
    "from dask.utils import parse_bytes\n",
    "from dask_cuda.device_host_file import DeviceHostFile\n",
    "\n",
    "path = \"/datasets/criteo/rzamora/crit_pq_int_2gb_split/ds_part.217.parquet\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Reading parquet from file (disk)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "233 ms ± 1.05 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%timeit cudf.io.read_parquet(path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Reading parquet from `BytesIO` (CPU memory) object"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 2.58 ms, sys: 251 ms, total: 253 ms\n",
      "Wall time: 252 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "mem_file = None\n",
    "with open(path, \"rb\") as f:\n",
    "    mem_file = BytesIO(f.read())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "241 ms ± 19.9 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%timeit cudf.io.read_parquet(mem_file)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### From Pandas (CPU memory) object"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 940 ms, sys: 1.1 s, total: 2.04 s\n",
      "Wall time: 2.04 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "pdf = cudf.io.read_parquet(path).to_pandas()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "418 ms ± 2.43 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%timeit cudf.from_pandas(pdf)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### From Arrow (CPU memory) object"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 535 ms, sys: 620 ms, total: 1.16 s\n",
      "Wall time: 1.15 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "adf = cudf.io.read_parquet(path).to_arrow()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "443 ms ± 5.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%timeit cudf.DataFrame.from_arrow(adf)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### From Pickled bytes (CPU memory) object"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 1.55 s, sys: 2.56 s, total: 4.11 s\n",
      "Wall time: 4.11 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "pdf = pickle.dumps(cudf.io.read_parquet(path))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.03 s ± 265 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%timeit pickle.loads(pdf)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Using `DeviceHostFile`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 499 ms, sys: 647 ms, total: 1.15 s\n",
      "Wall time: 1.15 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "dhf = DeviceHostFile(\n",
    "    device_memory_limit=parse_bytes(\"0GB\"),\n",
    "    memory_limit=parse_bytes(\"128GB\"),\n",
    "    local_directory=\"/raid/dask_space/rzamora/scratch\",\n",
    ")\n",
    "dhf[0] = cudf.io.read_parquet(path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "360 ms ± 1.83 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%timeit dhf.get(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}