First try boundary nudging

Fix range bug in TreeCompress, update parser to autofail non-singular…

.gitignore

@@ -1,4 +1,5 @@
 data.*
 __pycache__
-tree.json
-compressed_tree.json
+*.json
+.DS_Store
+.ipynb_checkpoints/

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "ideal-rmt-simulator"]
+	path = ideal-rmt-simulator
+	url = https://github.com/rfchang/ideal-rmt-simulator

CompressedTreeParser.ipynb

@@ -0,0 +1,128 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "938dec51",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "import pandas as pd\n",
+    "import argparse\n",
+    "from sklearn.tree import DecisionTreeClassifier, plot_tree, _tree\n",
+    "from sklearn.metrics import accuracy_score\n",
+    "from sklearn.tree import export_graphviz\n",
+    "import pydotplus\n",
+    "from matplotlib import pyplot as plt\n",
+    "from labels import mac_to_label\n",
+    "import json\n",
+    "import math"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "442624c7",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "Set1 = pd.read_csv('data.csv').values.tolist()\n",
+    "X = [i[0:3] for i in Set1]\n",
+    "Y =[i[3] for i in Set1]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "12ad454d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "predict_Yt = []\n",
+    "with open('compressed_tree.json', 'r') as file:\n",
+    "    data = json.load(file)\n",
+    "    classes = data[\"classes\"]\n",
+    "    for x in X:\n",
+    "        counter = 0\n",
+    "        class_set = []\n",
+    "        for feature in reversed(data['layers']): #Have to reverse this list due to structure of the data.csv file and how it aligns with the compressed tree layers\n",
+    "            for node in data['layers'][feature]:\n",
+    "                if node['min'] is None:\n",
+    "                    if x[counter] < node['max']:\n",
+    "                        class_set.append(node['classes'])\n",
+    "                        break #is this an issue?\n",
+    "                    else:\n",
+    "                        continue\n",
+    "                elif node['max'] is None:\n",
+    "                    if node['min'] < x[counter]:\n",
+    "                        class_set.append(node['classes'])\n",
+    "                        break #is this an issue?\n",
+    "                    else:\n",
+    "                        continue\n",
+    "                elif node['min'] < x[counter] and x[counter] < node['max']:\n",
+    "                    class_set.append(node['classes'])\n",
+    "                    break #is this an issue?\n",
+    "\n",
+    "            counter += 1\n",
+    "        result = set(class_set[0])\n",
+    "        for s in class_set[1:]:\n",
+    "            result.intersection_update(s)\n",
+    "\n",
+    "        #predict_Yt.append(list(result))\n",
+    "        #print(result)\n",
+    "        if len(result) == 1:\n",
+    "            prediction = list(result)[0]\n",
+    "            pred_class = classes[prediction]\n",
+    "            predict_Yt.append(pred_class)\n",
+    "        else:\n",
+    "            predict_Yt.append(None)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "8b4c56b6",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "0.8448217242194891\n"
+     ]
+    }
+   ],
+   "source": [
+    "correct = 0\n",
+    "for i in range(len(Y)):\n",
+    "    prediction = predict_Yt[i]\n",
+    "    if prediction != None and Y[i] == prediction:\n",
+    "        correct += 1\n",
+    "\n",
+    "print(correct / len(Y))"
+   ]
+  }
+ ],
+ "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": "3.12.7"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

ExtractDataset.ipynb

@@ -89,7 +89,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "switch",
+   "display_name": "Python 3 (ipykernel)",
    "language": "python",
    "name": "python3"
   },
@@ -103,7 +103,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.12.7"
+   "version": "3.12.9"
   }
  },
  "nbformat": 4,

TreeCompress.ipynb

@@ -2,7 +2,7 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": 9,
+   "execution_count": 1,
    "id": "ec310f34",
    "metadata": {},
    "outputs": [],
@@ -14,7 +14,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": 2,
    "id": "5b54797e",
    "metadata": {},
    "outputs": [],
@@ -28,22 +28,25 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 11,
+   "execution_count": 3,
    "id": "a38fdb8a",
    "metadata": {},
    "outputs": [],
    "source": [
     "# First cleanup the tree by rounding the decision points to integer values\n",
     "# We assume all features will use integer values. If this is not the case, then training data should be normalized so that integer values can be accurate enough\n",
-    "# we also enumerate all the paths for later use\n",
-    "\n",
     "i = 0\n",
     "\n",
+    "path_ids = set()\n",
+    "path_classes = tree[\"classes\"]\n",
+    "\n",
     "# for each path in the tree\n",
     "for path in paths:\n",
     "\t# assign a path id \n",
     "\tpath[\"id\"] = i\n",
-    "\ti += 1\n",
+    "\tpath_ids.add(i)\n",
+    "\t#path_classes.add(path[\"classification\"])\n",
+    "\ti += 1\t\n",
     "\t# for each condition\n",
     "\tconditions = path[\"conditions\"]\n",
     "\tfor condition in conditions:\n",
@@ -57,7 +60,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 4,
    "id": "2fd4f738",
    "metadata": {},
    "outputs": [],
@@ -80,7 +83,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": 5,
    "id": "98cde024",
    "metadata": {},
    "outputs": [],
@@ -106,13 +109,13 @@
     "\t\tvalue = condition[\"value\"]\n",
     "\n",
     "\t\t# move the min/max for the corresponding feature in compressed\n",
-    "\t\tif operation == \"<=\" and compressed[feature][\"min\"] is None:\n",
+    "\t\tif operation == \"<=\" and compressed[feature][\"max\"] is None:\n",
     "\t\t\tcompressed[feature][\"max\"] = value\n",
-    "\t\telif operation == \">\" and compressed[feature][\"max\"] is None:\n",
+    "\t\telif operation == \">\" and compressed[feature][\"min\"] is None:\n",
     "\t\t\tcompressed[feature][\"min\"] = value\n",
-    "\t\telif operation == \"<=\" and value < compressed[feature][\"min\"]:\n",
+    "\t\telif operation == \"<=\" and value < compressed[feature][\"max\"]:\n",
     "\t\t\tcompressed[feature][\"max\"] = value\n",
-    "\t\telif operation == \">\" and value > compressed[feature][\"max\"]:\n",
+    "\t\telif operation == \">\" and value > compressed[feature][\"min\"]:\n",
     "\t\t\tcompressed[feature][\"min\"] = value\n",
     "\n",
     "\tpath[\"compressed\"] = compressed"
@@ -120,7 +123,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": 6,
    "id": "b6fbadbf",
    "metadata": {},
    "outputs": [],
@@ -168,49 +171,58 @@
     "\t\t# for each bucket which encompases the condition's range, add this path's id to the sets \n",
     "\t\ti = 0\n",
     "\t\tfor bp in breakpoints[feature_name]:\n",
-    "\t\t\tin_range = is_in_range(bp, lower, upper)\n",
-    "\t\t\tif in_range:\n",
+    "\t\t\tif is_in_range(bp, lower, upper):\n",
     "\t\t\t\tbuckets_id[feature_name][i].add(ID)\n",
     "\t\t\t\tbuckets_class[feature_name][i].add(Class)\n",
-    "\t\t\ti += 1"
+    "\t\t\ti += 1\n",
+    "\n",
+    "\t\tif is_in_range(bp+1, lower, upper):\n",
+    "\t\t\tbuckets_id[feature_name][i].add(ID)\n",
+    "\t\t\tbuckets_class[feature_name][i].add(Class)"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 15,
+   "execution_count": 7,
    "id": "0a767971",
    "metadata": {},
    "outputs": [],
    "source": [
     "# combine breakpoints and buckets to one representation\n",
     "\n",
-    "compressed_tree = defaultdict(list)\n",
+    "compressed_layers = defaultdict(list)\n",
     "for feature_name in buckets_id:\n",
     "\tlower = None\n",
     "\tupper = breakpoints[feature_name][0]\n",
     "\tpaths = buckets_id[feature_name][0]\n",
     "\tclasses = buckets_class[feature_name][0]\n",
     "\t#print(f\"{feature_name} = [{lower}, {upper}]: {members}\")\n",
-    "\tcompressed_tree[feature_name].append({\"min\": lower, \"max\": upper, \"paths\": paths, \"classes\": classes})\n",
+    "\tcompressed_layers[feature_name].append({\"min\": lower, \"max\": upper, \"paths\": paths, \"classes\": classes})\n",
     "\tfor i in range(1, len(buckets_id[feature_name]) - 1):\n",
     "\t\tlower = breakpoints[feature_name][i-1]\n",
     "\t\tupper = breakpoints[feature_name][i]\n",
-    "\t\tmembers = buckets_id[feature_name][i]\n",
+    "\t\tpaths = buckets_id[feature_name][i]\n",
     "\t\tclasses = buckets_class[feature_name][i]\n",
     "\t\t#print(f\"{feature_name} = [{lower}, {upper}]: {buckets[feature_name][i]}\")\n",
-    "\t\tcompressed_tree[feature_name].append({\"min\": lower, \"max\": upper, \"paths\": paths, \"classes\": classes})\n",
+    "\t\tcompressed_layers[feature_name].append({\"min\": lower, \"max\": upper, \"paths\": paths, \"classes\": classes})\n",
     "\tlower = breakpoints[feature_name][len(breakpoints[feature_name]) - 1]\n",
     "\tupper = None\n",
-    "\tmembers = buckets_id[feature_name][len(buckets_id[feature_name]) - 1]\n",
+    "\tpaths = buckets_id[feature_name][len(buckets_id[feature_name]) - 1]\n",
     "\tclasses = buckets_class[feature_name][len(buckets_class[feature_name]) - 1]\n",
     "\t#print(f\"{feature_name} = [{lower}, {upper}]: {members}\")\n",
-    "\tcompressed_tree[feature_name].append({\"min\": lower, \"max\": upper, \"paths\": paths, \"classes\": classes})\n",
-    "\t#print(\"=\"*40)"
+    "\tcompressed_layers[feature_name].append({\"min\": lower, \"max\": upper, \"paths\": paths, \"classes\": classes})\n",
+    "\t#print(\"=\"*40)\n",
+    "\n",
+    "compressed_tree = {\n",
+    "\t\"paths\": path_ids,\n",
+    "\t\"classes\": path_classes,\n",
+    "\t\"layers\": compressed_layers,\n",
+    "}"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 16,
+   "execution_count": 8,
    "id": "561b0bc1",
    "metadata": {},
    "outputs": [],
@@ -229,7 +241,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "switch",
+   "display_name": "Python 3 (ipykernel)",
    "language": "python",
    "name": "python3"
   },
@@ -243,7 +255,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.12.7"
+   "version": "3.12.9"
   }
  },
  "nbformat": 4,

TreeToRMT.ipynb

@@ -0,0 +1,404 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "58fc6db9",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "import math"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "e07be4b3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "f = open(\"compressed_tree.json\")\n",
+    "tree = json.loads(f.read())\n",
+    "layers = tree[\"layers\"]\n",
+    "classes = tree[\"classes\"]\n",
+    "f.close()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "1516ff91",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "field_width = {\n",
+    "\t\"src\": 16,\n",
+    "\t\"dst\": 16,\n",
+    "\t\"protocl\": 8,\n",
+    "}"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "f9193827",
+   "metadata": {},
+   "source": [
+    "# Worst Case RMT"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "5e37cfc5",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def worst_case_rmt(tree):\n",
+    "\trmt = []\n",
+    "\tstep = 0\n",
+    "\n",
+    "\ttcam_bits = 0\n",
+    "\tram_bits = 0\n",
+    "\n",
+    "\tfor layer in layers:\n",
+    "\t\tnum_ranges = len(layers[layer])\n",
+    "\t\t# assume that each range requires all of 2*k prefixes when performing prefix expansion\n",
+    "\t\t# therefore there are 2*k * R for R ranges and width k\n",
+    "\t\tnum_prefixes = 2 * field_width[layer] * num_ranges\n",
+    "\t\tprefix_width = field_width[layer]\n",
+    "\n",
+    "\t\ttcam = {\n",
+    "\t\t\t\"id\": f\"{layer}_range\",\n",
+    "\t\t\t\"step\": step,\n",
+    "\t\t\t\"match\": \"ternary\",\n",
+    "\t\t\t\"entries\": num_prefixes,\n",
+    "\t\t\t\"key_size\": prefix_width\n",
+    "\t\t}\n",
+    "\t\ttcam_bits += num_prefixes * prefix_width\n",
+    "\n",
+    "\t\t# assume basic pointer reuse for metadata storage\n",
+    "\t\tram = {\n",
+    "\t\t\t\"id\": f\"{layer}_meta\",\n",
+    "\t\t\t\"step\": step,\n",
+    "\t\t\t\"match\": \"exact\",\n",
+    "\t\t\t\"method\": \"index\",\n",
+    "\t\t\t\"key_size\": math.ceil(math.log2(num_ranges)),\n",
+    "\t\t\t\"data_size\": len(classes)\n",
+    "\t\t}\n",
+    "\t\tram_bits += num_ranges * len(classes)\n",
+    "\n",
+    "\t\trmt.append(tcam)\n",
+    "\t\trmt.append(ram)\n",
+    "\n",
+    "\t\tstep += 1\n",
+    "\n",
+    "\treturn rmt, tcam_bits, ram_bits\n",
+    "\n",
+    "x, tcam_bits, ram_bits = worst_case_rmt(tree)\n",
+    "f = open(\"worst_case_rmt.json\", \"w+\")\n",
+    "f.write(json.dumps(x, indent=4))\n",
+    "f.close()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "0dc1d6d4",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "TCAM mapping: \n",
+      "[1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]\n",
+      "SRAM mapping: \n",
+      "[1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]\n",
+      "id mapping: \n",
+      "[['dst_range', 'dst_meta'], ['src_range', 'src_meta'], ['protocl_range', 'protocl_meta'], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], []]\n",
+      "TCAM bits: 13184\n",
+      "RAM bits:  504\n"
+     ]
+    }
+   ],
+   "source": [
+    "! command python3 ideal-rmt-simulator/sim.py naive_rmt.json\n",
+    "print(f\"TCAM bits: {tcam_bits}\")\n",
+    "print(f\"RAM bits:  {ram_bits}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "2a628655",
+   "metadata": {},
+   "source": [
+    "# Naive Range Expansion "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "id": "fb9febe9",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# shamelessly stolen from: https://github.com/autolyticus/range-to-prefix/blob/master/rangetoprefix.C\n",
+    "\n",
+    "def int_to_bin(i, width):\n",
+    "\treturn bin(i)[2:].zfill(width)\n",
+    "\n",
+    "def increment_dc(pfx):\n",
+    "\tidx = pfx.find(\"*\")\n",
+    "\tif idx == -1:\n",
+    "\t\tidx = len(pfx)\n",
+    "\tidx = idx - 1\n",
+    "\t#print(pfx, pfx[:idx])\n",
+    "\treturn pfx[:idx] + \"*\" + pfx[idx+1:]\n",
+    "\t\n",
+    "def can_merge(pfx_a, pfx_b):\n",
+    "\tpfx_a = pfx_a.replace(\"*\", \"\")\n",
+    "\tpfx_b = pfx_b.replace(\"*\", \"\")\n",
+    "\treturn pfx_a[:-1] == pfx_b[:-1] and pfx_a[-1] != pfx_b[-1]\n",
+    "\n",
+    "def merge(pfx_a, prefixes):\n",
+    "\tpfx_a = increment_dc(pfx_a)\n",
+    "\tprefixes[-1] = pfx_a\n",
+    "\n",
+    "\tfor i in range(len(prefixes) - 2, -1, -1):\n",
+    "\t\tif can_merge(prefixes[i], prefixes[i+1]):\n",
+    "\t\t\tprefixes.pop()\n",
+    "\t\t\tpfx = increment_dc(prefixes[i])\n",
+    "\t\t\tprefixes[i] = pfx\n",
+    "\n",
+    "def convert_range(lower, upper, width):\n",
+    "\tprefixes = []\n",
+    "\tprefix = int_to_bin(lower, width)\n",
+    "\tprefixes.append(prefix)\n",
+    "\tnorm_upper = min(upper, 2**width-1)\n",
+    "\tfor i in range(lower+1, norm_upper+1):\n",
+    "\t\tprefix = int_to_bin(i, width)\n",
+    "\t\tif can_merge(prefix, prefixes[-1]):\n",
+    "\t\t\tmerge(prefix, prefixes)\n",
+    "\t\telse:\n",
+    "\t\t\tprefixes.append(prefix)\n",
+    "\treturn prefixes"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "id": "55167c28",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def naive_rmt(tree):\n",
+    "\trmt = []\n",
+    "\tstep = 0\n",
+    "\n",
+    "\ttcam_bits = 0\n",
+    "\tram_bits = 0\n",
+    "\n",
+    "\tfor layer in layers:\n",
+    "\t\tnum_prefixes = 0\n",
+    "\t\tprefix_width = field_width[layer]\n",
+    "\t\t# for each range in the layer, convert the ranges to prefixes using naive range expansion\n",
+    "\t\tfor r in layers[layer]:\n",
+    "\t\t\tif r[\"min\"] == None:\n",
+    "\t\t\t\tr[\"min\"] = 0\n",
+    "\t\t\telif r[\"max\"] == None:\n",
+    "\t\t\t\tr[\"max\"] = 2 ** prefix_width\n",
+    "\t\t\tprefixes = convert_range(r[\"min\"], r[\"max\"], prefix_width)\n",
+    "\t\t\tr[\"prefixes\"] = prefixes\n",
+    "\t\t\tnum_prefixes += len(prefixes)\n",
+    "\t\t\ttcam_bits += len(prefixes) * prefix_width\n",
+    "\n",
+    "\t\ttcam = {\n",
+    "\t\t\t\"id\": f\"{layer}_range\",\n",
+    "\t\t\t\"step\": step,\n",
+    "\t\t\t\"match\": \"ternary\",\n",
+    "\t\t\t\"entries\": num_prefixes,\n",
+    "\t\t\t\"key_size\": prefix_width,\n",
+    "\t\t\t\"ranges\": layers[layer]\n",
+    "\t\t}\n",
+    "\n",
+    "\t\tnum_ranges = len(layers[layer])\n",
+    "\t\t# assume no pointer reuse for metadata storage\n",
+    "\t\tram = {\n",
+    "\t\t\t\"id\": f\"{layer}_meta\",\n",
+    "\t\t\t\"step\": step,\n",
+    "\t\t\t\"match\": \"exact\",\n",
+    "\t\t\t\"method\": \"index\",\n",
+    "\t\t\t\"key_size\": math.ceil(math.log2(num_ranges)),\n",
+    "\t\t\t\"data_size\": len(classes)\n",
+    "\t\t}\n",
+    "\t\tram_bits += num_ranges * len(classes)\n",
+    "\n",
+    "\t\trmt.append(tcam)\n",
+    "\t\trmt.append(ram)\n",
+    "\n",
+    "\t\tstep += 1\n",
+    "\n",
+    "\treturn rmt, tcam_bits, ram_bits\n",
+    "\n",
+    "x, tcam_bits, ram_bits = naive_rmt(tree)\n",
+    "f = open(\"naive_rmt.json\", \"w+\")\n",
+    "f.write(json.dumps(x, indent=4))\n",
+    "f.close()\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "id": "48011528",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "TCAM mapping: \n",
+      "[1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]\n",
+      "SRAM mapping: \n",
+      "[1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]\n",
+      "id mapping: \n",
+      "[['dst_range', 'dst_meta'], ['src_range', 'src_meta'], ['protocl_range', 'protocl_meta'], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], []]\n",
+      "TCAM bits: 3320\n",
+      "RAM bits:  504\n"
+     ]
+    }
+   ],
+   "source": [
+    "! command python3 ideal-rmt-simulator/sim.py naive_rmt.json\n",
+    "print(f\"TCAM bits: {tcam_bits}\")\n",
+    "print(f\"RAM bits:  {ram_bits}\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "id": "64b7271e",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# for this technique, we note that given disjoint ranges [0,a][a,b],[b,c] ...\n",
+    "# then if using a TCAM that selects the first matching prefix, then [0,a],[0,b],[0,c] would be equivalent\n",
+    "# this is because if for some k<a, even though the range [0,b] could be selected, as long as the prefixes for [0,a] are before [0,b] then the correct prefix will still be selected\n",
+    "\n",
+    "def priority_aware(tree):\n",
+    "\trmt = []\n",
+    "\tstep = 0\n",
+    "\n",
+    "\ttcam_bits = 0\n",
+    "\tram_bits = 0\n",
+    "\n",
+    "\tfor layer in layers:\n",
+    "\t\tnum_prefixes = 0\n",
+    "\t\tprefix_width = field_width[layer]\n",
+    "\t\t# for each range, run the regular prefix expansion, and also the prefix expansion setting the minimum to 0\n",
+    "\t\t# then check which set of prefixes would be better\n",
+    "\t\t# we will assume the ranges are already disjoin and in the correct order\n",
+    "\t\tfor r in layers[layer]:\n",
+    "\t\t\tif r[\"min\"] == None:\n",
+    "\t\t\t\tr[\"min\"] = 0\n",
+    "\t\t\telif r[\"max\"] == None:\n",
+    "\t\t\t\tr[\"max\"] = 2 ** prefix_width\n",
+    "\t\t\tregular_prefixes = convert_range(r[\"min\"], r[\"max\"], prefix_width)\n",
+    "\t\t\tzero_start_prefixes = convert_range(0, r[\"max\"], prefix_width)\n",
+    "\n",
+    "\t\t\tif len(regular_prefixes) <= len(zero_start_prefixes):\n",
+    "\t\t\t\tpfx_type = \"exact\"\n",
+    "\t\t\t\tprefixes = regular_prefixes\n",
+    "\t\t\telse:\n",
+    "\t\t\t\tpfx_type = \"zero\"\n",
+    "\t\t\t\tprefixes = zero_start_prefixes\n",
+    "\n",
+    "\t\t\tr[\"prefixes\"] = prefixes\n",
+    "\t\t\tr[\"prefix_type\"] = pfx_type\n",
+    "\t\t\tnum_prefixes += len(prefixes)\n",
+    "\t\t\ttcam_bits += len(prefixes) * prefix_width\n",
+    "\n",
+    "\t\ttcam = {\n",
+    "\t\t\t\"id\": f\"{layer}_range\",\n",
+    "\t\t\t\"step\": step,\n",
+    "\t\t\t\"match\": \"ternary\",\n",
+    "\t\t\t\"entries\": num_prefixes,\n",
+    "\t\t\t\"key_size\": prefix_width,\n",
+    "\t\t\t\"ranges\": layers[layer]\n",
+    "\t\t}\n",
+    "\n",
+    "\t\tnum_ranges = len(layers[layer])\n",
+    "\t\t# assume no pointer reuse for metadata storage\n",
+    "\t\tram = {\n",
+    "\t\t\t\"id\": f\"{layer}_meta\",\n",
+    "\t\t\t\"step\": step,\n",
+    "\t\t\t\"match\": \"exact\",\n",
+    "\t\t\t\"method\": \"index\",\n",
+    "\t\t\t\"key_size\": math.ceil(math.log2(num_ranges)),\n",
+    "\t\t\t\"data_size\": len(classes)\n",
+    "\t\t}\n",
+    "\t\tram_bits += num_ranges * len(classes)\n",
+    "\n",
+    "\t\trmt.append(tcam)\n",
+    "\t\trmt.append(ram)\n",
+    "\n",
+    "\t\tstep += 1\n",
+    "\n",
+    "\treturn rmt, tcam_bits, ram_bits\n",
+    "\n",
+    "x, tcam_bits, ram_bits = priority_aware(tree)\n",
+    "f = open(\"priority_aware.json\", \"w+\")\n",
+    "f.write(json.dumps(x, indent=4))\n",
+    "f.close()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 10,
+   "id": "cd706e41",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "TCAM mapping: \n",
+      "[1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]\n",
+      "SRAM mapping: \n",
+      "[1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]\n",
+      "id mapping: \n",
+      "[['dst_range', 'dst_meta'], ['src_range', 'src_meta'], ['protocl_range', 'protocl_meta'], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], []]\n",
+      "TCAM bits: 2152\n",
+      "RAM bits:  504\n"
+     ]
+    }
+   ],
+   "source": [
+    "! command python3 ideal-rmt-simulator/sim.py priority_aware.json\n",
+    "print(f\"TCAM bits: {tcam_bits}\")\n",
+    "print(f\"RAM bits:  {ram_bits}\")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "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": "3.12.9"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}