{ "cells": [ { "cell_type": "markdown", "metadata": { "hide_input": true }, "source": [ "# OPNFV VSPERF\n", "# Beyond Performance Metrics: Towards Causation Analysis" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### sridhar.rao@spirent.com and acm@research.att.com" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Import packages\n", "import numpy as np\n", "import pandas as pd\n", "import matplotlib.pyplot as plt\n", "import seaborn as sns\n", "from graphviz import Digraph\n", "import collections\n", "import glob\n", "import os" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Get the results to analyze: \n", "Getting Latest one, if ``directory_to_download`` is empty" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "import paramiko\n", "import tarfile\n", "import os\n", "from stat import S_ISDIR\n", "RECV_BYTES = 4096\n", "hostname = '10.10.120.24'\n", "port = 22\n", "uname='opnfv'\n", "pwd='opnfv' \n", "stdout_data = []\n", "stderr_data = []\n", "client = paramiko.Transport((hostname, port))\n", "client.connect(username=uname, password=pwd)\n", "session = client.open_channel(kind='session')\n", "directory_to_download = ''\n", "\n", "session.exec_command('ls /tmp | grep results')\n", "if not directory_to_download:\n", " while True:\n", " if session.recv_ready():\n", " stdout_data.append(session.recv(RECV_BYTES))\n", " if session.recv_stderr_ready():\n", " stderr_data.append(session.recv_stderr(RECV_BYTES))\n", " if session.exit_status_ready():\n", " break\n", " if stdout_data:\n", " line = stdout_data[0]\n", " filenames = line.decode(\"utf-8\").rstrip('\\n').split('\\n')\n", " filenames = sorted(filenames)\n", " latest = filenames[-1]\n", " directory_to_download = os.path.join('/tmp', latest).replace(\"\\\\\",\"/\")\n", " print(directory_to_download)\n", "stdout_data = []\n", "stderr_data = []\n", "if directory_to_download:\n", " # zip the collectd results to make the download faster\n", " zip_command = 'sudo -S tar -czvf '+ directory_to_download + '/collectd.tar.gz -C ' + '/tmp/csv .'\n", " session = client.open_channel(kind='session')\n", " session.get_pty()\n", " session.exec_command(zip_command)\n", " while True:\n", " if session.recv_ready():\n", " stdout_data.append(session.recv(RECV_BYTES))\n", " if session.recv_stderr_ready():\n", " stderr_data.append(session.recv_stderr(RECV_BYTES))\n", " if session.exit_status_ready():\n", " break\n", " if stderr_data:\n", " print(stderr_data[0])\n", " if stdout_data:\n", " print(stdout_data[0])\n", "\n", " # Begin the actual downlaod\n", " sftp = paramiko.SFTPClient.from_transport(client)\n", " def sftp_walk(remotepath):\n", " path=remotepath\n", " files=[]\n", " folders=[]\n", " for f in sftp.listdir_attr(remotepath):\n", " if S_ISDIR(f.st_mode):\n", " folders.append(f.filename)\n", " else:\n", " files.append(f.filename)\n", " if files:\n", " yield path, files\n", " # Filewise download happens here\n", " for path,files in sftp_walk(directory_to_download):\n", " for file in files:\n", " remote = os.path.join(path,file).replace(\"\\\\\",\"/\")\n", " local = os.path.join('/data/results', file).replace(\"\\/\",\"/\")\n", " sftp.get(remote, local)\n", "# Untar the collectd results if we got it.\n", "path = os.path.join('/data/results', 'collectd.tar.gz')\n", "if os.path.exists(path):\n", " tar = tarfile.open(path)\n", " tar.extractall()\n", " tar.close()\n", "# Ready to work with downloaded data, close the session and client.\n", "session.close()\n", "client.close()" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "strings = ('* OS:', '* Kernel Version:', '* Board:', '* CPU:', '* CPU cores:',\n", " '* Memory:', '* Virtual Switch Set-up:',\n", " '* Traffic Generator:','* vSwitch:', '* DPDK Version:', '* VNF:')\n", "filename = os.path.basename(glob.glob('/data/results/result*.rst')[0])\n", "info_dict = {}\n", "with open(os.path.join('/data/results', filename), 'r') as file:\n", " for line in file:\n", " if any(s in line for s in strings):\n", " info_dict[line.split(':', 1)[0]] = line.split(':', 1)[1].rstrip()\n", "df = pd.DataFrame.from_dict(info_dict, orient='index', columns=['Value'])\n", "df" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Understand the configuration used for the test." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "filename = os.path.basename(glob.glob('/data/results/vsperf*.conf')[0])\n", "file = os.path.join('/data/results', filename)\n", "with open(file, 'r') as f:\n", " for line in f:\n", " if line.startswith('TRAFFICGEN_DURATION'):\n", " value = line.split('=')[1]\n", " value = value.rstrip()\n", " value = value.lstrip()\n", " traffic_duration = int(value)\n", " print(traffic_duration)\n", " elif line.startswith('VSWITCH_PMD_CPU_MASK'):\n", " value = line.split('=')[1]\n", " value = value.rstrip()\n", " pmd_cores_mask = value.lstrip()\n", " print(pmd_cores_mask)\n", " elif line.startswith('GUEST_CORE_BINDING'):\n", " value = line.split('=')[1]\n", " value = value.rstrip()\n", " value = value.lstrip()\n", " guest_cores = value[1:-2]\n", " print(guest_cores)" "\n", ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## OVS-Ports and Cores" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "import collections\n", "portcores = collections.OrderedDict()\n", "chunks = []\n", "current_chunk = []\n", "file = os.path.join('/data/results', 'ovs-cores.log')\n", "with open(file, 'r') as f:\n", " for line in f:\n", " if line.startswith('pmd') and current_chunk:\n", " # if line starts with token and the current chunk is not empty\n", " chunks.append(current_chunk[:]) # add not empty chunk to chunks\n", " current_chunk = [] # make current chunk blank\n", " # just append a line to the current chunk on each iteration\n", " if \"port:\" in line or 'pmd' in line:\n", " current_chunk.append(line)\n", " chunks.append(current_chunk) # append the last chunk outside the loop\n", "\n", "core_ids = []\n", "for ch in chunks:\n", " port_id = ''\n", " core_id = ''\n", " for line in ch:\n", " if 'pmd' in line:\n", " core_id = line.split()[-1][:-1]\n", " if core_id not in core_ids:\n", " core_ids.append(core_id)\n", " elif 'port:' in line:\n", " port_id = line.split()[1]\n", " if port_id and core_id:\n", " if port_id not in portcores:\n", " portcores[port_id] = core_id\n", "\n", "# import graphviz\n", "from graphviz import Digraph\n", "ps = Digraph(name='ovs-ports-cores', node_attr={'shape': 'box'}, edge_attr={'arrowhead':\"none\"})\n", "with ps.subgraph(name=\"cluster_0\") as c:\n", " c.node_attr.update(style='filled', color='green')\n", " c.node('t0', 'TGen-Port-0')\n", " c.node('t1', 'TGen-Port-1')\n", " c.attr(label='TGEN')\n", " c.attr(color='blue')\n", "with ps.subgraph(name=\"cluster_1\") as c:\n", " c.node_attr.update(style='filled', color='yellow')\n", " c.node('v0', 'VNF-Port-0')\n", " c.node('v1', 'VNF-Port-1')\n", " c.attr(label='VNF')\n", " c.attr(color='blue')\n", " \n", "with ps.subgraph(name='cluster_2') as c: \n", " c.attr(label='OVS-DPDK')\n", " c.attr(color='blue')\n", " count = 0\n", " for port, core in portcores.items():\n", " id = 'o'+str(count)\n", " c.node(id, port+'\\nCore-ID:'+ core)\n", " count += 1\n", " num = port[-1]\n", " if 'dpdkvhost' in port:\n", " ps.edge(id, 'v'+num)\n", " else:\n", " ps.edge(id, 't'+num)\n", "\n", "ps" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Dropped Packets" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "portcores = collections.OrderedDict()\n", "chunks = []\n", "current_chunk = []\n", "file = os.path.join('/data/results', 'ovs-cores.log')\n", "with open(file, 'r') as f:\n", " for line in f:\n", " if line.startswith('pmd') and current_chunk:\n", " # if line starts with token and the current chunk is not empty\n", " chunks.append(current_chunk[:]) # add not empty chunk to chunks\n", " current_chunk = [] # make current chunk blank\n", " # just append a line to the current chunk on each iteration\n", " if \"port:\" in line or 'pmd' in line:\n", " current_chunk.append(line)\n", " chunks.append(current_chunk) # append the last chunk outside the loop\n", "\n", "core_ids = []\n", "for ch in chunks:\n", " port_id = ''\n", " core_id = ''\n", " for line in ch:\n", " if 'pmd' in line:\n", " core_id = line.split()[-1][:-1]\n", " if core_id not in core_ids:\n", " core_ids.append(core_id)\n", " elif 'port:' in line:\n", " port_id = line.split()[1]\n", " if port_id and core_id:\n", " if port_id not in portcores:\n", " portcores[port_id] = core_id\n", "\n", "ps = Digraph(name='ovs-dropped', node_attr={'shape': 'box'}, edge_attr={'arrowhead':\"none\"})\n", "\n", "def get_dropped(port_id):\n", " # port_id = 'dpdk0'\n", " if glob.glob('./pod12-node4/*'+port_id):\n", " dirname = os.path.basename(glob.glob('./pod12-node4/*'+port_id)[0])\n", " if dirname:\n", " if glob.glob('./pod12-node4/'+dirname+ '/*dropped*'):\n", " filename = os.path.basename(glob.glob('./pod12-node4/'+dirname+ '/*dropped*')[0])\n", " if filename:\n", " with open(os.path.join('./pod12-node4', dirname, filename), 'r') as f:\n", " line = f.readlines()[-1]\n", " fields = line.split(',')\n", " return fields[1], fields[2]\n", " return 'NA','NA'\n", "\n", "with ps.subgraph(name=\"cluster_0\") as c:\n", " c.node_attr.update(style='filled', color='pink')\n", " c.attr(label='OVS-DPDK')\n", " c.attr(color='blue')\n", " count = 0\n", " for port, core in portcores.items():\n", " id = 'o'+str(count)\n", " rx,tx = get_dropped(port)\n", " c.node(id, port+'\\nRX-Dropped:'+ rx + '\\nTX-Dropped:' + tx)\n", " count += 1\n", " num = port[-1]\n", "ps" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Plotting Live Results - T-Rex" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "lines_seen = set() # holds lines already seen\n", "outfile = open('./counts.dat', \"w\")\n", "file = os.path.join('/data/results', 'trex-liveresults-counts.dat')\n", "for line in open(file, \"r\"):\n", " if line not in lines_seen: # not a duplicate\n", " outfile.write(line)\n", " lines_seen.add(line)\n", "outfile.close()\n", "tdf = pd.read_csv('./counts.dat')\n", "print(tdf.columns)\n", "ax = tdf.loc[(tdf.rx_port == 1)].plot(y='rx_pkts')\n", "def highlight(indices,ax):\n", " i=0\n", " while i int(traffic_duration):\n", " highlight([(ind, i)], ax)\n", " ind = i\n", " indv = tdf.ts[i]\n", "highlight([(ind,i)], ax)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## IRQ Latency Histogram" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "file = os.path.join('/data/results', 'RUNirq.irq.log')\n", "tdf = pd.read_csv(file)\n", "tdf.columns\n", "exclude = [' <1', ' < 5', ' < 10',' < 50', ' < 100', ' < 500', ' < 1000']\n", "ax = tdf.loc[:, tdf.columns.difference(exclude)].plot(x=' number', xticks=tdf[' number'], figsize=(20,10))\n", "ax.set_xlabel('Core #')\n", "ax.set_ylabel('Count')\n", "#tdf.plot(x='number')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Sample Collectd Metric Display - L3 Cache Occupancy in Bytes" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import math\n", "def cpumask2coreids(mask):\n", " intmask = int(mask, 16)\n", " i = 1\n", " coreids = []\n", " while (i < intmask):\n", " if (i & intmask):\n", " coreids.append(str(math.frexp(i)[-1]-1))\n", " i = i << 1\n", " return (coreids)\n", "\n", "vswitch_cpus = \"['2']\"\n", "ps = Digraph(name='cpu-map', node_attr={'shape': 'box'}, edge_attr={'arrowhead':\"none\"})\n", "with ps.subgraph(name=\"cluster_0\") as c:\n", " c.node_attr.update(style='filled', color='pink')\n", " c.attr(label='CPU-MAPPINGS')\n", " c.attr(color='blue')\n", " c.node('vscpus', 'vSwitch: \\n' + vswitch_cpus)\n", " # vnf_cpus = cpumask2coreids(guest_cores)\n", " c.node('vncpus', 'VNF: \\n' + guest_cores)\n", " pmd_cpus = cpumask2coreids(pmd_cores_mask[1:-1])\n", " c.node('pmcpus', 'PMDs: \\n' + str(pmd_cpus))\n", "\n", "ps" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "# Path where collectd results are stored.\n", "mypath = \"./pod12-node4\"\n", "file_count = 0\n", "cpu_names = []\n", "for level1 in os.listdir(mypath):\n", " if \"intel_rdt\" in level1:\n", " l2path = os.path.join(mypath, level1)\n", " for level2 in os.listdir(l2path):\n", " if \"bytes\" in level2:\n", " l3path = os.path.join(l2path, level2)\n", " if file_count == 0:\n", " file_count += 1\n", " df = pd.read_csv(l3path)\n", " nn = 'cpu-'+ level1[len('intel_rdt-'):]\n", " # nn = 'cpu-'+ level1.split('-')[1]\n", " cpu_names.append(nn)\n", " # print(nn)\n", " df.rename(columns={'value': nn}, inplace=True)\n", " else:\n", " file_count += 1\n", " tdf = pd.read_csv(l3path)\n", " nn = 'cpu-'+ level1[len('intel_rdt-'):]\n", " cpu_names.append(nn)\n", " tdf.rename(columns={'value': nn}, inplace=True)\n", " df[nn] = tdf[nn] \n", "\n", "ax = df.plot(x='epoch', y=cpu_names)\n", "ax.set_ylabel(\"MBytes\")\n", "ax.set_xlabel('Time')\n", "\n", "\n", " \n", "# df = pd.read_csv()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Events " ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "from datetime import datetime\n", "filename = os.path.basename(glob.glob('/data/results/vsperf-overall*.log')[0])\n", "logfile = os.path.join('/data/results', filename)\n", "linecnt = 0\n", "times = {}\n", "with open(logfile) as f:\n", " for line in f:\n", " line = line.strip('\\n')\n", " if linecnt == 0:\n", " times['Start-Test'] = line.split(\" : \")[0]\n", " linecnt += 1\n", " if 'Binding NICs' in line:\n", " times['Binding-NICs'] = line.split(\" : \")[0]\n", " if 'Starting traffic at' in line:\n", " sline = line.split(\" : \")[1]\n", " time = line.split(\" : \")[0]\n", " speed = sline.split('at',1)[1]\n", " times[speed] = time \n", " elif 'Starting vswitchd' in line:\n", " times['vSwitch-Start'] = line.split(\" : \")[0]\n", " elif 'Starting ovs-vswitchd' in line:\n", " times['ovsvswitch-start'] = line.split(\" : \")[0]\n", " elif 'Adding Ports' in line:\n", " times['Ports-Added'] = line.split(\" : \")[0]\n", " elif 'Flows Added' in line:\n", " times['Flows-Added'] = line.split(\" : \")[0]\n", " elif 'send_traffic with' in line:\n", " times['Traffic Start'] = line.split(\" : \")[0]\n", " elif 'l2 framesize 1280' in line:\n", " times['Traffic-Start-1280'] = line.split(\" : \")[0]\n", " elif 'Starting qemu' in line:\n", " times['VNF-Start'] = line.split(\" : \")[0]\n", " elif 'l2 framesize 64' in line:\n", " times['Traffic-Start-64'] = line.split(\" : \")[0]\n", " elif 'l2 framesize 128' in line:\n", " times['Traffic-Start-128'] = line.split(\" : \")[0]\n", " elif 'l2 framesize 256' in line:\n", " times['Traffic-Start-256'] = line.split(\" : \")[0]\n", " elif 'l2 framesize 512' in line:\n", " times['Traffic-Start-512'] = line.split(\" : \")[0]\n", " elif 'l2 framesize 1024' in line:\n", " times['Traffic-Start-1024'] = line.split(\" : \")[0]\n", " elif 'l2 framesize 1518' in line:\n", " times['Traffic-Start-1518'] = line.split(\" : \")[0]\n", " elif 'dump flows' in line:\n", " times['Traffic-End'] = line.split(\" : \")[0]\n", " elif 'Wait for QEMU' in line:\n", " times['VNF-Stop'] = line.split(\" : \")[0]\n", " elif 'delete flow' in line:\n", " times['flow-removed'] = line.split(\" : \")[0]\n", " elif 'delete port' in line:\n", " times['port-removed'] = line.split(\" : \")[0]\n", " elif 'Killing ovs-vswitchd' in line:\n", " times['vSwitch-Stop'] = line.split(\" : \")[0]\n", "\n", "times['Test-Stop'] = line.split(\" : \")[0]\n", "#print(times)\n", "ddf = pd.DataFrame.from_dict(times, orient='index', columns=['timestamp'])\n", "names = ddf.index.values\n", "dates = ddf['timestamp'].tolist()\n", "datefmt=\"%Y-%m-%d %H:%M:%S,%f\"\n", "dates = [datetime.strptime(ii, datefmt) for ii in dates]\n", "# print(names)\n", "# print(dates)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "import matplotlib.dates as mdates\n", "from matplotlib import ticker\n", "\n", "levels = np.array([-5, 5, -3, 3, -1, 1])\n", "fig, ax = plt.subplots(figsize=(40, 5))\n", "\n", "# Create the base line\n", "start = min(dates)\n", "stop = max(dates)\n", "ax.plot((start, stop), (0, 0), 'k', alpha=.5)\n", "\n", "pos_list = np.arange(len(dates))\n", "\n", "# Iterate through releases annotating each one\n", "for ii, (iname, idate) in enumerate(zip(names, dates)):\n", " level = levels[ii % 6]\n", " vert = 'top' if level < 0 else 'bottom'\n", " ax.scatter(idate, 0, s=100, facecolor='w', edgecolor='k', zorder=9999)\n", " # Plot a line up to the text\n", " ax.plot((idate, idate), (0, level), c='r', alpha=.7)\n", " # Give the text a faint background and align it properly\n", " ax.text(idate, level, iname,\n", " horizontalalignment='right', verticalalignment=vert, fontsize=14,\n", " backgroundcolor=(1., 1., 1., .3))\n", "ax.set(title=\"VSPERF Main Events\")\n", "# Set the xticks formatting\n", "ax.get_xaxis().set_major_locator(mdates.SecondLocator(interval=30))\n", "ax.get_xaxis().set_major_formatter(mdates.DateFormatter(\"%M %S\"))\n", "fig.autofmt_xdate()\n", "plt.setp((ax.get_yticklabels() + ax.get_yticklines() +\n", " list(ax.spines.values())), visible=False)\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Current and old." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Current Result" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "import glob\n", "filename = os.path.basename(glob.glob('/data/results/result*.csv')[0])\n", "filename\n", "tdf = pd.read_csv(os.path.join('/data/results', filename))\n", "pkts = ['tx_frames', 'rx_frames']\n", "fps = ['tx_rate_fps', 'throughput_rx_fps']\n", "mbps = ['tx_rate_mbps', 'throughput_rx_mbps']\n", "pcents = ['tx_rate_percent', 'throughput_rx_percent', 'frame_loss_percent']\n", "fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(14, 12))\n", "tdf.plot.bar(y= pkts,ax=axes[0,0])\n", "tdf.plot.bar(y= fps,ax=axes[0,1])\n", "tdf.plot.bar(y= mbps,ax=axes[1,0])\n", "tdf.plot.bar(y= pcents,ax=axes[1,1])\n", "current_pkt_size = str(tdf['packet_size'].iloc[-1])\n", "current_rx_fps = str(tdf['throughput_rx_fps'].iloc[-1])\n", "print(current_rx_fps)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## How Current Result compares to Previous ones?" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "hide_input": true }, "outputs": [], "source": [ "import urllib\n", "import json\n", "import requests\n", "#json_data = requests.get('http://testresults.opnfv.org/test/api/v1/results?project=vsperf').json()\n", "json_data = requests.get('http://10.10.120.22:8000/api/v1/results?project=vsperf').json()\n", "res = json_data['results']\n", "df1 = pd.DataFrame(res)\n", "sort_by_date = df1.sort_values('start_date')\n", "details = df1['details'].apply(pd.Series)\n", "details[current_pkt_size] = pd.to_numeric(pd.Series(details[current_pkt_size]))\n", "# details.plot.bar(y = current_pkt_size)\n", "details_cur_pkt = details[[current_pkt_size]].copy()\n", "details_cur_pkt.loc[-1]= float(current_rx_fps)\n", "details_cur_pkt.index = details_cur_pkt.index + 1 # shifting index\n", "details_cur_pkt.sort_index(inplace=True) \n", "ax = details_cur_pkt.plot.bar()\n", "ax.set_ylabel(\"Frames per sec\")\n", "ax.set_xlabel(\"Run Number\")\n", "def highlight(indices,ax):\n", " i=0\n", " while i