5.24. Measuring performance of provisioning systems

status

ready

version

1.0

Abstract

This document describes a test plan for quantifying the performance of provisioning systems as a function of the number of nodes to be provisioned. The plan includes the collection of several resource utilization metrics, which will be used to analyze and understand the overall performance of each system. In particular, resource bottlenecks will either be fixed, or best practices developed for system configuration and hardware requirements.

Conventions

  • Provisioning: is the entire process of installing and configuring an operating system.

  • Provisioning system: is a service or a set of services which enables the installation of an operating system and performs basic operations such as configuring network interfaces and partitioning disks. A preliminary list of provisioning systems can be found below in Applications. The provisioning system can include configuration management systems like Puppet or Chef, but this feature will not be considered in this document. The test plan for configuration management systems is described in the “Measuring_performance_of_configuration_management_systems” document.

  • Performance of a provisioning system: is a set of metrics which describes how many nodes can be provisioned at the same time and the hardware resources required to do so.

  • Nodes: are servers which will be provisioned.

5.24.1. Test Plan

This test plan aims to identify the best provisioning solution for cloud deployment, using specified list of performance measurements and tools.

5.24.1.1. Test Environment

5.24.1.1.1. Preparation

  1. The following package needs to be installed on the provisioning system servers to collect performance metrics.

Software to be installed

package name

version

source

dstat

0.7.2

Ubuntu trusty universe repository

5.24.1.1.2. Environment description

Test results MUST include a description of the environment used. The following items should be included:

  • Hardware configuration of each server. If virtual machines are used then both physical and virtual hardware should be fully documented. An example format is given below:

Description of server hardware

server

name

role

vendor,model

operating_system

CPU

vendor,model

processor_count

core_count

frequency_MHz

RAM

vendor,model

amount_MB

NETWORK

interface_name

vendor,model

bandwidth

STORAGE

dev_name

vendor,model

SSD/HDD

size

  • Configuration of hardware network switches. The configuration file from the switch can be downloaded and attached.

  • Configuration of virtual machines and virtual networks (if used). The configuration files can be attached, along with the mapping of virtual machines to host machines.

  • Network scheme. The plan should show how all hardware is connected and how the components communicate. All ethernet/fibrechannel and VLAN channels should be included. Each interface of every hardware component should be matched with the corresponding L2 channel and IP address.

  • Software configuration of the provisioning system. sysctl.conf and any other kernel file that is changed from the default should be attached. List of installed packages should be attached. Specifications of the operating system, network interfaces configuration, and disk partitioning configuration should be included. If distributed provisioning systems are to be tested then the parts that are distributed need to be described.

  • Desired software configuration of the provisioned nodes. The operating system, disk partitioning scheme, network interface configuration, installed packages and other components of the nodes affect the amount of work to be performed by the provisioning system and thus its performance.

5.24.1.2. Test Case

5.24.1.2.1. Description

This specific test plan contains only one test case, that needs to be run step by step on the environments differing list of parameters below.

5.24.1.2.2. Parameters

Parameter name

Value

number of nodes

10, 20, 40, 80, 160, 320, 640, 1280, 2000

5.24.1.2.3. List of performance metrics

The table below shows the list of test metrics to be collected. The priority is the relative ranking of the importance of each metric in evaluating the performance of the system.

List of performance metrics

Priority

Value

Measurement Units

Description

1

PROVISIONING_TIME

seconds
The elapsed time to provision all
nodes, as a function of the numbers of
nodes

2

INGRESS_NET

Gbit/s
Incoming network bandwidth usage as a
function of the number of nodes.
Average during provisioning on the host
where the provisioning system is
installed.

2

EGRESS_NET

Gbit/s
Outgoing network bandwidth usage as a
function of the number of nodes.
Average during provisioning on the host
where the provisioning system is
installed.

3

CPU

percentage
CPU utilization as a function of the
number of nodes. Average during
provisioning on the host where the
provisioning system is installed.

3

RAM

GB
Active memory usage as a function of
the number of nodes. Average during
provisioning on the host where the
provisioning system is installed.

3

WRITE_IO

operations/second
Storage read IO bandwidth as a
function of the number of nodes.
Average during provisioning on the host
where the provisioning system is
installed.

3

READ_IO

operations/second
Storage write IO bandwidth as a
function of the number of nodes.
Average during provisioning on the host
where the provisioning system is
installed.

5.24.1.2.4. Measuring performance values

The script Full script for collecting performance metrics can be used for the first five of the following steps.

Note

If a distributed provisioning system is used, the values need to be measured on each provisioning system instance.

  1. Start the collection of CPU, memory, network, and storage metrics during the provisioning process. Use the dstat programm which can collect all of these metrics in CSV format into a log file.

  2. Start the provisioning process for the first node and record the wall time.

  3. Wait until the provisioning process has finished (when all nodes are reachable via ssh) and record the wall time.

  4. Stop the dstat program.

  5. Prepare collected data for analysis. dstat provides a large amount of information, which can be pruned by saving only the following:

    • “system”[time]. Save as given.

    • 100-“total cpu usage”[idl]. dstat provides only the idle CPU value. CPU utilization is calculated by subtracting the idle value from 100%.

    • “memory usage”[used]. dstat provides this value in Bytes. This is converted it to Megabytes by dividing by 1024*1024=1048576.

    • “net/eth0”[recv] receive bandwidth on the NIC. It is converted to Megabits per second by dividing by 1024*1024/8=131072.

    • “net/eth0”[send] send bandwidth on the NIC. It is converted to Megabits per second by dividing by 1024*1024/8=131072.

    • “net/eth0”[recv]+”net/eth0”[send]. The total receive and transmit bandwidth on the NIC. dstat provides these values in Bytes per second. They are converted to Megabits per second by dividing by 1024*1024/8=131072.

    • “io/total”[read] storage read IO bandwidth.

    • “io/total”[writ] storage write IO bandwidth.

    • “io/total”[read]+”io/total”[writ]. The total read and write storage IO bandwidth.

    These values will be graphed and maximum values reported.

    Additional tests will be performed if some anomalous behaviour is found. These may require the collection of additional performance metrics.

  6. The result of this part of test will be:

  • to provide the following graphs, one for each number of provisioned nodes:

    1. Three dependencies on one graph.

      • INGRESS_NET(TIME) Dependence on time of incoming network bandwidth usage.

      • EGRESS_NET(TIME) Dependence on time of outgoing network bandwidth usage.

      • ALL_NET(TIME) Dependence on time of total network bandwidth usage.

    2. One dependence on one graph.

      • CPU(TIME) Dependence on time of CPU utilization.

    3. One dependence on one graph.

      • RAM(TIME) Dependence on time of active memory usage.

    4. Three dependencies on one graph.

      • WRITE_IO(TIME) Dependence on time of storage write IO bandwidth.

      • READ_IO(TIME) Dependence on time of storage read IO bandwidth.

      • ALL_IO(TIME) Dependence on time of total storage IO bandwidth.

Note

If a distributed provisioning system is used, the above graphs should be provided for each provisioning system instance.

  • to fill in the following table for maximum values:

The resource metrics are obtained from the maxima of the corresponding graphs above. The provisioning time is the elapsed time for all nodes to be provisioned. One set of metrics will be given for each number of provisioned nodes.

Maximum values of performance metrics
nodes
count
provisioning
time
maximum
CPU
usage
maximum
RAM
usage
maximum
NET
usage
maximum
IO
usage

10

20

40

80

160

320

640

1280

2000

5.24.2. Applications

5.24.2.1. List of provisioning systems

list of provisioning systems

Name of provisioning system

Version

Cobbler

2.4

Razor

0.13

Image based provisioning via downloading images with bittorrent protocol

5.24.3. Full script for collecting performance metrics

#!/bin/bash

# Need to install the required packages on provisioning system servers:
if (("`dpkg -l | grep dstat | grep ^ii > /dev/null; echo $?` == 1"))
then
  apt-get -y install dstat
fi

# Need to prepare the following script on provisioning system server to collect
# values of CPU,RAM,NET and IO loads per second. You need to change "INTERFACE"
# variable regarding the interface which connected to nodes to communicare with
# them during provisioning process. As a result of this command we'll get
# running in backgroud dstat programm which collecting needed parametes in CSV
# format into /var/log/dstat.log file.:
INTERFACE=eth0
OUTPUT_FILE=/var/log/dstat.csv
dstat --nocolor --time --cpu --mem --net -N ${INTERFACE} --io --output ${OUTPUT_FILE} > /dev/null &

# Need to prepare script which starts provisioning process and gets the time when
# provisioning started and when provisioning ended ( when all nodes reachable via
# ssh). We'll analyze results collected during this time window. For getting
# start time we can add "date" command before API call or CLI command and forward
# the output of the command to some log file. Here is example for cobbler:
ENV_NAME=env-1
start_time=`date +%s.%N`
echo "Provisioning started at "`date` > /var/log/provisioning.log
for SYSTEM in `cobbler system find --comment=${ENV_NAME}`
do
  cobbler system reboot --name=$i &
done

# For getting end-time we can use the script below. This script tries to reach
# nodes via ssh and write "Provisioning finished at <date/time>" into
# /var/log/provisioning.log file. You'll need to provide ip addresses of the
# nodes (from file nodes_ips.list, where IPs listed one per line) and
# creadentials (SSH_PASSWORD and SSH_USER variables):
SSH_OPTIONS="StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null"
SSH_PASSWORD="r00tme"
SSH_USER="root"
NODE_IPS=(`cat nodes_ips.list`)
TIMER=0
TIMEOUT=20
while (("${TIMER}" < "${TIMEOUT}"))
do
     for NODE_IP in ${NODE_IPS[@]}
     do
             SSH_CMD="sshpass -p ${SSH_PASSWORD} ssh -o ${SSH_OPTIONS} ${SSH_USER}@${NODE_IP}"
             ${SSH_CMD} "hostname" && UNHAPPY_SSH=0 || UNHAPPY_SSH=1
             if (("${UNHAPPY_SSH}" == "0"))
             then
                     echo "Node with ip "${NODE_IP}" is reachable via ssh"
                     NODE_IPS=(${NODE_IPS[@]/${NODE_IP}})
             else
                     echo "Node with ip "${NODE_IP}" is still unreachable via ssh"
             fi
      done
      TIMER=$((${TIMER} + 1))
      if (("${TIMER}" == "${TIMEOUT}"))
      then
              echo "The following "${#NODE_IPS[@]}" are unreachable"
              echo ${NODE_IPS[@]}
              exit 1
      fi
      if ((${#NODE_IPS[@]} == 0 ))
      then
              break
      fi
      # Check that nodes are reachable once per 1 seconds
      sleep 1
done
echo "Provisioning finished at "`date` > /var/log/provisioning.log

end_time=`date +%s.%N`
elapsed_time=$(echo "$end_time - $start_time" | bc -l)
echo "Total elapsed time for provisioning: $elapsed_time seconds" > /var/log/provisioning.log

# Stop dstat command
killall dstat

# Delete excess values and convert to needed metrics. So, we'll get the
# following csv format:
# time,cpu_usage,ram_usage,net_recv,net_send,net_all,dsk_io_read,dsk_io_writ,dsk_all
awk -F "," 'BEGIN {getline;getline;getline;getline;getline;getline;getline;
                   print "time,cpu_usage,ram_usage,net_recv,net_send,net_all,dsk_io_read,dsk_io_writ,dsk_all"}
            {print $1","100-$4","$8/1048576","$12/131072","$13/131072","($12+$13)/131072","$14","$15","$14+$15}' \
$OUTPUT_FILE > /var/log/10_nodes.csv

5.24.4. Reports

Test plan execution reports: