CN106301954A - A kind of method of Smart Rack grid test optimization - Google Patents

Abstract

The present invention provides a kind of method of Smart Rack grid test optimization, relates to Smart Rack and network field, the invention mainly comprises two aspects and is optimized, and first, by configuring system service, completion system environment optimization arranges and installs testing tool；Secondly, whole machine cabinet resource being divided into two groups, by accurate proportioning, half system resource carries out netserver, and half system resource carries out netperf and test process is accurately bundled in fixing CPU core tests.Solve Smart Rack in the problem that network performance is not enough and handling capacity is the most up to standard towards the storage of mass data with when processing, can apply in cloud computing, big data process etc. with better services.

Description

A kind of method of Smart Rack grid test optimization

Technical field

The present invention relates to Smart Rack and network field, particularly relate to a kind of Smart Rack grid test optimization Method.

Background technology

Netperf is the survey tool of a kind of network performance, mainly for transmission based on TCP or UDP.Netperf root According to the difference of application, the applied in network performance test of different mode, i.e. bulk data transfer pattern and request/response mould can be carried out Formula.What Netperf test result was reflected is that a system can send data with speed how soon to another one system, with And another one system can be with speed receiving data how soon.

When building or manage a network system, will often find that, in actual network environment uses, web application Using effect less desirable, problem possibly be present at the exploitation of program above, it is also possible to due to actual network environment In there is bottleneck.The five indices weighing network performance is:

1. availability

The first step of testing characteristics of network determines that network the most normally works, and simplest method is to use ping order. By to far-end machine send icmp echo request, and etc. icmp echo reply to be received judge far-end Machine whether connect, network the most normally works.Network device internal typically has multiple Buffer Pool, different Buffer Pools to use Different buffer size, is respectively intended to process different size of packet (packet).

2. response time

It is exactly response time the cost time that the echo request/reply of Ping order once comes and goes institute.

3. network utilization

Network utilization refers to that the time that network is used accounts for total time the ratio of (time of the time i.e. used+free time). Therefore, network test tool generally uses network throughput and network bandwidth capacity to the property determining in network between two nodes Energy.

4. network throughput

Network throughput refers to sometime, between two nodes in a network, it is provided that to the remaining bandwidth of network application.

5. network bandwidth capacity

Different from network throughput, network bandwidth capacity refers to the maximum available bandwidth between two nodes of network.This It is to be determined by the ability of the equipment of network consisting.

Smart Rack is with standard cabinet as designing unit, and its server node carried is powered, dispels the heat by unification And management, it is widely used in the fields such as big data cloud calculating, and its network performance always affects its handling capacity, causes system Performance cannot play, to such an extent as to serious waste, such as cannot access connected reference and respond slow without response or system.

Summary of the invention

In order to solve problem above, the present invention proposes a kind of method of Smart Rack grid test optimization.

The side of Smart Rack grid test maneuver optimization is mainly by configuring system service completion system environment Optimal design-aside, by accurate proportioning, carries out netserver by system resource half, and half carries out netperf by test process Accurately it is tied to CPU core and is optimized test.

The optimization of 1 configuration system service completion system environmental variable

Prohibit to fall irqbanlace and cpuspeed and can obtain the network throughput of maximum to obtain best test effect.

1.1. forbid affecting the system service of network:

/sbin/service cpuspeed start

/sbin/service irqbalance stop

chkconfig irqbalance off

chkconfig cpuspeed on

1.2. complete the installation of testing tool and whole machine cabinet system resource is configured to two groups.

cd /opt/scripts;tar -zxvf netperf-2.4.5.tar.gz;cd netperf-2.4.5;./ configure;

make;make install

for i in `seq 1 96`

do

num=$(((i-1)/32+1))

j=$(((i-1)%32+1))

if [ $j -lt 10 ] ; then

j=0${j}

fi

Test process is accurately tied to CPU core by 2, and system resource half is carried out netserver, and half carries out netperf Test.

Netperf instrument works in client/server mode.Server end is netserver, be used for intercept from The connection of client end, client end is netperf, is used for initiating network test to server.Client Yu server it Between, by accurate matching system resource, it is allowed to each process and can exclusively enjoy its allocated resource, it is to avoid various processes are done back and forth Disturb or system resource switches over.

2.1. available CPU core resource be divided into two groups in statistics node.

proc=`cat /proc/cpuinfo |grep processor |wc -l`

num=$(($proc/2))

for i in `seq 0 $(($num-1))`

Do

2.2. the CPU core of wherein half is bound port.

port=$((12887+$i))

taskset -c $i /usr/local/bin/netserver -p $port >/dev/null 2>1

taskset -pc 2 $! >/dev/null

Done

2.3. inspection netperf connects IP port.

for i in `seq 0 $(($num-1))`

do

port=$((12887+$i))

echo --- Netperf Connect to $ip $port

while [ "`/usr/bin/nmap $ip -p $port |grep $port |awk '{print $2}'`" != "open" >/dev/null ]

do

sleep 5

done

Done

2.4. the netperf test being optimized.

echo Begin Test............

for i in `seq $num $(($proc-1))`

do

port=$((12888+$i-$num))

taskset -c $i /usr/local/bin/netperf -H $ip -l $2 -p $port >>/ tmp/${biosserial}/netperf 2>1 &

taskset -pc $i $! /dev/null 2>1

done

The invention has the beneficial effects as follows

Optimized by precision net netperf performance, network test can be prevented effectively from two-way to running out of existing speed instability very To the problem being decreased obviously, accurately bound by process-core, maximum resource utilization can be avoided the wave of resource during test Take, solve Smart Rack not enough and the most up to standard asking of handling capacity at network performance towards the storage of mass data and when processing Topic, by the Smart Rack network optimization, is allowed to better services and applies in cloud computing, big data process etc..

Detailed description of the invention

Below present disclosure is carried out more detailed elaboration:

The method of Smart Rack grid test optimization includes two broad aspect: by configuring system service, completion system ring Border optimal design-aside also installs testing tool；Carrying out accurate proportioning by whole machine cabinet resource is divided into two groups, half system resource is entered Row netserver, half system resource carries out netperf and test process is accurately bundled in fixing CPU core tests. Optimized by precision net netperf performance, network test can be prevented effectively from two-way the brightest to running out of existing speed instability The aobvious problem declined, is accurately bound by process-core, and maximum resource utilization can be avoided the waste of resource during test.

1, the optimization of system service completion system environmental variable is configured

1.1. forbid affecting the system service of network, prohibit to fall irqbanlace and cpuspeed to obtain maximum network throughput Amount；

/sbin/service cpuspeed start；

/sbin/service irqbalance stop；

chkconfig irqbalance off；

chkconfig cpuspeed on。

1.2. complete the installation of testing tool and whole machine cabinet system resource is configured to two groups.

2, test process being accurately tied to CPU core, system resource half is carried out netserver, half is carried out Netperf tests.

2.1. available CPU core resource be divided into two groups in statistics node；

2.2. the CPU core of wherein half is bound port；

2.3. inspection netperf connects IP port；

2.4. the netperf test being optimized.

Claims (3)

1. the method for a Smart Rack grid test optimization, it is characterised in that

Being optimized including two aspects, first, by configuring system service, completion system environment optimization arranges and installs test Instrument；Secondly, whole machine cabinet resource being divided into two groups, by accurate proportioning, half system resource carries out netserver, half system System resource carries out netperf and test process is accurately bundled in fixing CPU core tests.

Method the most according to claim 1, it is characterised in that

The optimization of configuration system service completion system environmental variable

Prohibit to fall irqbanlace and cpuspeed

1). forbid affecting the system service of network:

/sbin/service cpuspeed start

/sbin/service irqbalance stop

chkconfig irqbalance off

chkconfig cpuspeed on

2). complete the installation of testing tool and whole machine cabinet system resource is configured to two groups；

cd /opt/scripts;tar -zxvf netperf-2.4.5.tar.gz;cd netperf-2.4.5;./ configure;

make;make install

for i in `seq 1 96`

do

num=$(((i-1)/32+1))

j=$(((i-1)%32+1))

if [ $j -lt 10 ] ; then

j=0${j}

fi。

Method the most according to claim 1, it is characterised in that

Test process being accurately tied to CPU core, system resource half is carried out netserver, half carries out netperf and enters Row test；

Netperf instrument works in client/server mode；Server end is netserver, is used for intercepting from client The connection of end, client end is netperf, is used for initiating network test to server；Between client and server, pass through Accurately matching system resource, is allowed to each process and can exclusively enjoy its allocated resource, it is to avoid various processes disturb back and forth or System resource switches over；

1). available CPU core resource be divided into two groups in statistics node

proc=`cat /proc/cpuinfo |grep processor |wc -l`

num=$(($proc/2))

for i in `seq 0 $(($num-1))`

Do

2). the CPU core of wherein half is bound port

port=$((12887+$i))

taskset -c $i /usr/local/bin/netserver -p $port >/dev/null 2>1

taskset -pc 2 $! >/dev/null

Done

3). inspection netperf connects IP port

for i in `seq 0 $(($num-1))`

do

port=$((12887+$i))

echo --- Netperf Connect to $ip $port

while [ "`/usr/bin/nmap $ip -p $port |grep $port |awk '{print $2}'`" != "open" >/dev/null ]

do

sleep 5

done

Done

4). the netperf test being optimized

echo Begin Test............

for i in `seq $num $(($proc-1))`

do

port=$((12888+$i-$num))

taskset -c $i /usr/local/bin/netperf -H $ip -l $2 -p $port >>/ tmp/${biosserial}/netperf 2>1 &

taskset -pc $i $! /dev/null 2>1

done。