Activity-15.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "import io_backend as io"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "**Note: this notebook requires `io_backend.py` and `compModel.js` to be in the same directory**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Sequential Flooding: An example of better buffer management for joins\n",
    "------------------------------------------------------------------\n",
    "\n",
    "Recall: A _buffer manager_ is a system to handle operations supporting the buffer\n",
    "* For example, a buffer manager should handle **eviction** of pages from the buffer when space needs to be freed up for reading in new pages.\n",
    "\n",
    "In general the OS has a built-in buffer manager; however, most DBMSs implement their own.  *Why?*"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "**Key concept:** Often, a DBMS can potentially implement *more efficient* buffer management since it knows what operations it is carrying out.\n",
    "\n",
    "## For Example: When Performing a Join!\n",
    "\n",
    "We'll observe a phenomenon known as **sequential flooding** that occurs when a buffer manager uses a naive default eviction policy while looping over data, e.g. in conditions that would occur when a DBMS was performing a join!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Exercise 1: NLJ in Python\n",
    "\n",
    "Just to get warmed up, write a nested loop join in python, that takes as inputs **two lists of dictionaries**, and joins them on the attribute `A`, returning tuples which have attribute `A` and the union of the other attributes:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# SAMPLE INPUT\n",
    "R = [{'A':0,'B':1},{'A':1,'B':1},{'A':2,'B':1}]\n",
    "S = [{'A':0,'C':3},{'A':3,'C':0},{'A':0,'C':5}]\n",
    "\n",
    "# Your join algorithm here\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Exercise 2: LRU eviction & sequential flooding\n",
    "\n",
    "In general, most OS buffer managers will have a **least recently used (LRU)** eviction policy, which means that the least recently used page\\* will be evicted if more space is needed in the buffer.\n",
    "\n",
    "\\* *[Note: here we define \"used\" to mean any sort of action involving the page, for simplicty!]*\n",
    "\n",
    "Let's see what happens under this policy when we have a buffer of size $B$ and want to loop over $B+1$ pages $M$ times:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# Create pages & flush to disk\n",
    "# NOTE: set display mark so that we don't bother displaying this\n",
    "def init_pages(b, n_pages):\n",
    "    file_id = b.new_file()\n",
    "    for i in range(n_pages):\n",
    "        page = b.new_page(file_id)\n",
    "        b.flush(page)\n",
    "    b.display_set_mark()\n",
    "    return file_id\n",
    "\n",
    "# Loop through one iteration of the file, highlighting the LRU/MRU page\n",
    "def get_next_page(b, fid, pid, eviction_method='LRU'):\n",
    "    try:\n",
    "        page = b.read(fid, pid)\n",
    "    except io.BufferMemoryException:\n",
    "        old_page, old_buffer_idx = b.get_buffer_page(eviction_method)\n",
    "        b.release(old_page)\n",
    "        page = b.read(fid, pid)\n",
    "    return page"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "We will do the following $M$ times:\n",
    "* For $i$ in range $B+1$:\n",
    "    * If page $i$ is already in buffer, read from buffer (*fast!*)\n",
    "    * Else:\n",
    "        * If buffer **is full**:\n",
    "            * **Evict**- i.e. flush to disk- the LRU page (*slow!*)\n",
    "        * Read page $i$ from disk -> buffer (*slow!*), then read from buffer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "EVICTION_METHOD = 'LRU'\n",
    "BUFFER_SIZE = 3\n",
    "N_PAGES = BUFFER_SIZE + 1\n",
    "M = 3\n",
    "\n",
    "# Initialize buffer\n",
    "b = io.Buffer(buffer_size=BUFFER_SIZE, page_size=1, buffer_queue_indicator=EVICTION_METHOD)\n",
    "file_id = init_pages(b, N_PAGES)\n",
    "\n",
    "# Do M ordered passes over the N pages\n",
    "for i in range(M):\n",
    "    for pid in range(N_PAGES):\n",
    "        page = get_next_page(b, file_id, pid, eviction_method=EVICTION_METHOD)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Can you see what happens?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# Visualize what we just did\n",
    "b.display(speed=1000)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "For general $M$ and $N>B$, what is the IO cost here?  And why is this the case?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Exercise 3: A better replacement policy?\n",
    "\n",
    "Can you think of a better replacement policy?  It should be comparabley simple, but lead to better IO cost for our scenario..."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Exercise 4: A better IO Cost...\n",
    "\n",
    "With this new replacement policy, what is the IO cost for $M$ and $N>B$ now?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "EVICTION_METHOD = # YOUR BETTER POLICY NAME HERE?\n",
    "BUFFER_SIZE = 3\n",
    "N_PAGES = BUFFER_SIZE + 1\n",
    "M = 3\n",
    "\n",
    "# Initialize buffer\n",
    "b = io.Buffer(buffer_size=BUFFER_SIZE, page_size=1, buffer_queue_indicator=EVICTION_METHOD)\n",
    "file_id = init_pages(b, N_PAGES)\n",
    "\n",
    "# Do M ordered passes over the N pages\n",
    "for i in range(M):\n",
    "    for pid in range(N_PAGES):\n",
    "        page = get_next_page(b, file_id, pid, eviction_method=EVICTION_METHOD)\n",
    "        \n",
    "# Visualize what we just did- note that a different buffer_num must be specified so it displays in own pane!\n",
    "b.display(speed=1000, buffer_num=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}