CN101533366A - Method for acquiring and analyzing performance data of server - Google Patents

Abstract

The invention provides a method for acquiring and analyzing performance data of a server. The method uses data acquisition and analysis as core, forms standardized operation through the sorting and standardization of an acquisition method and a target, acquires key data capable of reflecting server performance, uses the data as evidence and analyzes each subsystem of the server, thereby finding the performance bottleneck of the server and providing evidence for solving test or promoting the server performance in practical application.

Description

The method of a kind of performance data of server collection and analysis

Technical field

The present invention relates to a kind of Computer Applied Technology field or relate to server application performance testing authentication field, the method for specifically a kind of performance data of server collection and analysis.

Background technology

No matter server product is that we need performance data is carried out collection analysis, thereby to the total system operation conditions is monitored, pinpoint the problems as early as possible in test or practical application when relating to concrete application.Especially descend in the system applies performance, when going wrong, we more need performance data is gathered, thereby find the problem place, deal with problems.

Summary of the invention

The method that the purpose of this invention is to provide a kind of performance data of server collection and analysis.

The objective of the invention is to realize in the following manner, concrete steps are as follows:

One, data acquisition scope

1) in test process, write down the test data in the whole test process,

2) determine the object of performance monitoring, comprise each subsystem, mainly comprise cpu, internal memory, disk, network;

Two, data acquisition synchronism

1) test macro may be divided into multi-layer framework, and the data aggregation on each layer architecture will guarantee temporal synchronous, could make accurately the state variation of system's run duration like this and describing;

2) the performance monitoring data of each subsystem will go up synchronously the assurance time;

3) test log: in test process, all will finish test log every day, writes down one day test process, goes wrong;

Three, collecting method: the collection of data according to the difference of platform, is divided into two classes: Linux and windows, wherein: linux system

1) performance monitoring instrument

A) cpu, internal memory: vmstat

B) magnetic disc i/o: iostat

C) network: sar

The Windows system

2) performance monitoring instrument:

In the following order, open performance monitor: control panel-management tool-performance

Four, test data arrangement: the arrangement of test data, the test data formization is perhaps graphical, for convenient follow-up analysis is prepared;

1) test data patternization: at linux system, the data that we collect under the linux system all are text formattings, need us these graphical datas, are convenient to follow-up data analysis contrast; Figure comprises broken line graph or histogram;

2) file designation rule:

A) allow filename comprise more fileinfo as far as possible;

B) form: instrument abbreviation _ platform information _ test model;

Five, method of testing:

1) sampling time interval: in once testing together, the sampling time interval of different instruments need equate;

2) adopt point value: sampled point is a numerical value of N,

A) test duration is below 1 hour: 150＜N＜250;

B) test duration is 1-2 hour: 250＜N＜450;

3) use of vmstat: at first need to create directory, be used for depositing test data:

[root@nodel~]#cd/tmp/

[root@nodel?tmp]#mkdir?test_vmstat

Carry out as issuing orders:

[root@nodel?tmp]#?vmstat?5n>/tmp/test_vmstat/file_name

# per 5 second value once, and the result outputed in the file;

4) use of iostat

At first need to create directory, be used for depositing test data:

[root@nodel~]#cd/tmp/

[root@nodel?tmp]#mkdir?test_iostat

Carry out as issuing orders:

[root@nodel?tmp]#iostat-x5n|grep?sdn>/tmp/test_iostat/file_name

# per 5 second value once, and the result outputed in the file;

5) use of sar

At first need to create directory, be used for depositing test data:

[root@nodel~]#cd/tmp/

[root@nodel?tmp]#mkdir?test_sar

Carry out as issuing orders:

[root@nodel?tmp]#sar-n?DEV5n|grep?eth0>/tmp/test_sar/file_name

# per 5 second value once, and the result outputed in the file;

Six, windows performance monitoring instrument: control panel-management tool-performance

Use step:

1) counter daily record: open the performance monitoring instrument, click performance daily record and alarm, this step is finished in the gated counter daily record, and the syslog file that an acquiescence is arranged is seen in the perform region on the right: system overview

2) newly-built daily record setting

Click by right key perform region on the right side, selects newly-built daily record setting, ejects dialog box, inserts Log Names;

3) add counter

Click and add the counter button, eject dialog box, in the performance object the inside, the object that selection will be monitored comprises processor, internal memory, network;

The corresponding some counters of each object are chosen gated counter from tabulation, after adding the counter that needs and finishing, carry out next step

4) journal file is set

Above step is conventional the setting, and the click logs file carries out the setting of journal file now;

Keep acquiescence herein;

5) plan is set

Be set to manually carry out or as required, plan to be provided with;

6) test beginning starts, and Icon Color has redness to become green;

Newly-built daily record is finished, and just sees newly-built daily record in counter daily record workspace; Red icon is a halted state, and green is a running status;

Choose daily record, click right or manual the execution begin and stop;

Six test datas are understood and are analyzed

About the Linux platform: in the test process, by vmstat, iostat, sar surveillance resource mainly comprises: process, internal memory, swap, cpu, magnetic disc i/o and network interface card;

Vmstat analyzes: key parameter:

1) process: we mainly pay close attention to r and b parameter:

The r parameter: the operation queue number of threads, comprise the thread that moving and the thread of waiting for CPU timeslice, this numeral has at least a thread to want waiting for CPU greater than the number of CPU, and the thread of waiting for CPU is many more, might exert an influence to performance more;

The b parameter: the kernel thread average in the per second waiting list, just comprise thread at the I/O of pending file system, or owing to the internal memory thread that control is suspended of packing into; Process is owing to the internal memory control of packing into is suspended, and the obstruction row (b) in vmstat report show that number of threads increases, rather than number of threads increases in the operation queue;

2) cpu: use the number percent segmentation of CPU time in this time interval, the cpu row are as follows:

Us parameter: us row have shown the CPU time that user model consumed;

Sy parameter: sy row have shown that in detail CPU carries out the number percent that thread is taken time under system model;

Id parameter: id row have shown the number percent of CPU free time when not having unsettled magnetic disc i/o or stand-by period, do not have unsettled I/O, and the time of the wait that is useful on was included in free time;

Wa parameter: wa row have shown the percentage of time of CPU free time when unsettled magnetic disc i/o is arranged in detail, has a uncompleted magnetic disc i/o when moving at least when waiting for, this time is included into the time of waiting for I/O, this process is used asynchronous I/O, then the I/O to disk asks invokes thread to be got clogged or sleep, and is done up to request; In case the I/O of process request is finished, this process is placed in the operation queue, and I/O finishes very soon, and this process is used more CPU time; The value of wa should not surpass 25%

3) other parameters

Memory

Swpd: virtual memory operating position, unit: KB;

Free: idle internal memory, unit K B

Buff: be used as the interior poke of buffer memory, unit: KB;

Swap

Si: exchange to the exchange number of pages amount of internal memory, unit: KB/ second from disk;

So: exchange to the exchange number of pages amount of disk, unit: KB/ second from internal memory;

IO

Bi: send to the piece number of block device, unit: piece/second;

Bo: from the piece number that block device receives, unit: piece/second;

Analyze:

1) r is much larger than the number of cpu, and system will be slack-off, and r is greater than 4 times of the cpu number, and illustrative system is faced with the cpu shortage of power, and system can be seriously slack-off;

2) the id value often is 0, and sy+us is often greater than 80%, and illustrative system cpu overload lacks the cpu resource in other words;

3) the wa value illustrates disk not by good balance greater than 25%, and system waits for that the load of magnetic disc i/o or disk is very heavy the cost plenty of time;

4) linux at first uses physical memory, re-uses the swap subregion after physical memory uses, thus linux system general memory utilization factor all than higher, si, so equal 0, illustrate that the swap subregion is not used;

Improvement project:

1) adjusts application, make its better utilization cpu, increase cpu number or rising cpu dominant frequency;

2) optimize disk, solve the I/O problem;

Iostat analyzes: key parameter:

RkB/s: per second is read the K byte number, is half of rsect/s, because every sector-size is 512 bytes;

WkB/s: per second is write the K byte number, is half of wsect/s;

Avgqu-sz: average I/O queue, IOQ length;

Await: the stand-by period (millisecond) of average each equipment I/O operation;

Svctm: the service time (millisecond) of average each equipment I/O operation;

%util: had percent what time to be used for the I/O operation in one second, have how long I/O queue, IOQ is a non-NULL in one second in other words;

Analyze:

1) disk shows high for a long time read-write, and svctm is also much larger than 30 simultaneously, and the value of await is also much larger than svctm, and there is serious magnetic disc i/o bottleneck in illustrative system so;

2) %util is always near 100%, and the I/O request that illustrative system produces is too many, and may there be disk bottleneck in system;

3) size of wait generally depends on svctm and the pattern of sending of the length of I/O queue, IOQ and I/O request service time, and svctm illustrates that relatively near await I/O does not almost have the stand-by period; Await illustrates that much larger than svctm I/O queue, IOQ is oversize, and it is slack-off to use the response time that obtains

Improvement project:

1) adjusts application, make it use disk i/o more efficient, by revising the cache of disk queue, use application server;

2) file system is distributed on 2 or a plurality of disk, changes disk raid scheme;

3) change better dish battle array or disk, use storage scheme faster;

About the windows platform

The windows platform test will be at the selected performance monitoring object in performance monitor the inside, and for each monitored object adds the performance monitoring counter, the counter that performance monitoring object and each object comprise is as follows then:

1) processor processor

Processor performance object comprises the counter of weighing processor activity aspect, processor be computing machine count and logical calculated, the associate member startup operation and the operation processing threads part, a computing machine has many processors, and handler object is with the example of every processor as object;

Mainly comprise counter:

%Processor Time-finger processor is used for carrying out the number percent of non-idle thread time, computing method are, measure the time of the at interval interior non-idle thread activity of example, deduct this value at interval with example, this value is that every processor has an idle thread, cycles consumed when this thread moves when not having other threads, this counter is the main explanation device of processor activity, be presented at example observed rush hour average percent at interval the time, this value deducts this with 100% and serves inactive Time Calculation and come out;

2) memory internal memory

Memory performance object is made up of the physics on the description computing machine and the counter of behavioral virtual memory, physical memory refers to the quantity of the random access memory on the computing machine, virtual memory is made up of the space on physical memory and the disk, many internal memory counters monitor paging, too much paging causes delay, can influence the total system treatment effeciency;

The counter that mainly comprises has:

Available MBytes is meant the physical memory amount of representing with MB, and this internal memory can distribute to a process at once or system uses, and it equals to distribute to the summation of standby, free time and zero page or leaf list memory;

3) Physical Disk hard disk

Physical Disk performance object comprises the hard disk on the supervisory computer or the counter of fixed disk drive, disk is used for storage file, program and paged data and by reading these projects of retrieval and writing by record it is changed, and the value of physical disk counter is the summation of logic magnetic disc value;

The counter that mainly comprises has:

Avg.Disk Queue Length refers to read and write the average of request;

%Disk Time refers to that selected disc driver is busy with providing the service number percent of used time for reading or writing request;

4) Network Interface network

Network Interface performance object comprises the counter of weighing by the speed of a TCP/IP network connection transmission and reception byte and packet, and it comprises the counter of supervising connection error;

The counter that mainly comprises has:

Bytes Total/sec is the speed that sends and receive byte on each network adapter, comprise the frame character, Network Interface Bytes Received/sec be Network Interface BytesReceived/sec and Network Interface the summation of Bytes Sent/sec;

Elaborate below, before the test beginning each time, preferably server and application are restarted, and guarantee not disturb between twice test, in the test process, need to carry out corresponding change according to concrete actual conditions.

The invention has the beneficial effects as follows: for the method for server performance collection, with data acquisition and analysis is core, by arrangement and standard to acquisition method and target, form normalizing operation, the critical data that can reflect server performance is gathered, and is foundation with data, and each subsystem of server is analyzed, thereby find the server performance bottleneck, the lifting server performance provides foundation in test or the practical application in order to solve.

Embodiment

1. data acquisition scope

1) data acquisition scope

2) in test process, write down the test data in the whole test process.

3) determine the object of performance monitoring, comprise each subsystem, mainly comprise cpu, internal memory, disk, network.

2 data acquisition synchronisms

1) test macro may be divided into multi-layer framework, and the data aggregation on each layer architecture will guarantee temporal synchronous, could make accurately the state variation of system's run duration like this and describing.

2) the performance monitoring data of each subsystem will go up synchronously the assurance time.

3) test log: in test process, all will finish test log every day, writes down one day test process, goes wrong.

3, collecting method

The collection of data according to the difference of platform, mainly is divided into two classes: Linux ﹠amp; Windows

Linux system

1) performance monitoring instrument

A) cpu, internal memory: vmstat

B) magnetic disc i/o: iostat

C) network: sar

2) using method (seeing 4 trifles)

The Windows system

1) performance monitoring instrument

In the following order, open performance monitor:

Control panel-〉 management tool-performance

2) using method (seeing 4 trifles)

4, test data arrangement

The arrangement of test data, the test data formization is perhaps graphical, for convenient follow-up analysis is prepared

1) test data patternization

Mainly be at linux system, the data that we collect under the linux system all are text formattings, need us these graphical datas, are convenient to follow-up data analysis contrast.

Mainly be that figure comprises broken line graph or histogram

2) file designation rule

In test process, the value file is many, its standardize naming need be avoided confusion.Be convenient to later use.

Naming rule:

A) allow filename comprise more fileinfo as far as possible

B) form: instrument abbreviation _ platform information _ test model (can change) according to actual conditions

For example: in the mis system test of Henan State Family Planning Commission, the following name of file:

Io_4N4U---4 nodes, 4 channel platforms, iostat test data

Vm_it---Itanium platform, the vmstat test data

Io_lk_r-1000 user iostat reads performance curve.

5, method of testing example

1. sampling time interval.

In once testing together, the sampling time interval of different instruments need equate.

2. adopt point value.Sampled point is a numerical value of N

For ease of the analysis of follow-up test, guarantee the sampled point of sufficient amount, advise according to the test duration:

A) test duration is below 1 hour: 150＜N＜250

B) test duration is 1-2 hour: 250＜N＜450

Vmstat uses

At first need to create directory, be used for depositing test data:

[root@nodel~]#cd/tmp/

[root@nodel?tmp]#mkdir?test_vmstat

Carry out as issuing orders:

[root@nodeltmp]#vmstat?5n>/tmp/test_vmstat/file_name

# per 5 second value once, and the result outputed in the file.

Note:1) file_name names according to actual conditions

2) n is the number of times of record, adjusts according to the test duration.

3) value interval time is 5 seconds, also can adjust.

Example：(vmstat)

[root@standby~]#vmstat15

procs--------memory-----------swap-------io-----system------cpu----

r b swpd free buff cache si?so bi bo in cs us sy?id wa

0 0 0 69680?60500?533600?0 0 1 0 52 16 0 0 100 0

0 0 0 69680?60500?533600?0 0 0 0 1008?16 0 0 100 0

0 0 0 69680?60500?533600?0 0 0 0 1025?35 0 0 100 0

0 0 0 69680?60500?533600?0 0 0 0 1004?17 0 0 100 0

0 0 0 69680?60500?533600?0 0 0 0 1026?35 0 0 100 0

Io stat uses

At first need to create directory, be used for depositing test data:

[root@nodel~]#cd/tmp/

[root@nodel?tmp]#?mkdir?test_iostat

Carry out as issuing orders:

[root@nodel?tmp]#iostat-x5n|grep?sdn>/tmp/test_iostat/file_name

# per 5 second value once, and the result outputed in the file.

Note:1) file_name names according to actual conditions

2) n is the number of times of record, adjusts according to the test duration.

3) value interval time is 5 seconds, also can adjust.

4) the disk unit title of sdn for monitoring revised according to actual conditions

Example：(iostat)

[root@standby~]#iostat-xll|grep?sdb

Device：?rrqm/s wrqm/s r/s w/s rsec/s wsec/s rkB/s

sdb 0.00 0.00 0.00 0.00?0.01 0.00 0.00

Device：?wkB/s avgrq-sz?avgqu-sz await svctm ％util

sdb 0.00 33.60 0.00 13.27 5.44 0.00

Sar uses

At first need to create directory, be used for depositing test data:

[root@nodel~]#cd/tmp/

[root@nodel?tmp]#?mkdir?test_sar

Carry out as issuing orders:

[root@nodel?tmp]#sar-n?DEV5n|grep?eth0>/tmp/test_sar/file_name

# per 5 second value once, and the result outputed in the file.

Note:1) file_name names according to actual conditions

2) n is the number of times of record, adjusts according to the test duration.

3) value interval time is 5 seconds, also can adjust.

Example：(sar)

[root@standby~]#sar-n?DEV?15|grep?ethl

11:25:10AM?IFACE?rxpck/s?txpck/s?rxbyt/s?txbyt/s?rxcmp/s?txcmp/s?rxmcst/s

11:25:10AM?ethl 2.00 0.00 120.00 0.00 0.00 0.00 0.00

11:25:11AM?ethl 0.98 0.98 58.82 119.61 0.00 0.00 0.00

11:25:12AM?ethl 6.00 1.00 580.00 122.00 0.00 0.00 0.00

11:25:13AM?ethl 3.00 1.00 212.00 106.00 0.00 0.00 0.00

11:25:14AM?ethl 3.00 1.00 180.00 106.00 0.00 0.00 0.00

Average: ethl 2.99 0.80 229.48 90.84 0.00 0.00 0.00

Windows performance monitoring instrument

Control panel-〉 management tool-performance

Use step:

1) counter daily record

Open the performance monitoring instrument, click performance daily record and alarm, the gated counter daily record

Finish this step, the syslog file that an acquiescence is arranged: systemoverview can be seen in the perform region on the right

2) newly-built daily record setting

Click by right key perform region on the right side, selects newly-built daily record setting, ejects dialog box, inserts Log Names.

3) add counter

Click and add the counter button, eject dialog box, in the performance object the inside, the object that selection will be monitored, such as processor, internal memory, network.

The corresponding some counters of each object can be chosen gated counter from tabulation, add the counter that needs, and after finishing, carry out next step

4) journal file is set

Above step is conventional the setting, and the click logs file carries out the setting of journal file now.

Can keep acquiescence herein.

5) plan is set

Be set to manual execution.Also can plan to be provided with as required.

6) test beginning starts, and Icon Color has redness to become green.

Newly-built daily record is finished, and just can see newly-built daily record in counter daily record workspace.Red icon is a halted state, and green is a running status.

Choose daily record, click right can manually be carried out beginning and stop.

6, test data is understood and is analyzed

The Linux platform

In the test process, we mainly pass through vmstat, iostat, and sar surveillance resource mainly comprises: process, internal memory, swap, cpu, magnetic disc i/o, and network interface card.

Vmstat analyzes

Key parameter:

1) process: we mainly pay close attention to r and b parameter:

The r parameter: the operation queue number of threads comprises the thread that moving and the thread of waiting for CPU timeslice.If this numeral greater than the number of CPU, has at least a thread to want waiting for CPU, the thread of waiting for CPU is many more, might exert an influence to performance more.

B parameter: the kernel thread average in the per second waiting list.Just comprise thread at the I/O of pending file system, or owing to the internal memory thread that control is suspended of packing into.

If process is owing to the internal memory control of packing into is suspended, the obstruction row (b) in vmstat report show that number of threads increases, rather than number of threads increases in the operation queue.

2) cpu: in this time interval, use the number percent segmentation of CPU time.The cpu row are as follows:

Us parameter: us row have shown the CPU time that user model consumed.

Sy parameter: sy row have shown that in detail CPU carries out the number percent that thread is taken time under system model.

Id parameter: id row have shown the number percent of CPU free time when not having unsettled magnetic disc i/o or stand-by period.If there is not unsettled I/O, the time of the wait that is useful on was included in free time.

Wa parameter: the percentage of time of CPU free time during dish I/O.If have at least a uncompleted magnetic disc i/o, this time to be included into the time of waiting for I/O when moving when waiting for.Unless this process is used asynchronous I/O, otherwise to the I/O request of disk make invokes thread get clogged (or sleep) be done up to request.In case the I/O of process request is finished, this process is placed in the operation queue.If I/O finishes very soon, this process can be used more CPU time.The value of wa should not surpass 25%

3) other parameters

Memory

Swpd: virtual memory operating position, unit: KB

Free: idle internal memory, unit K B

Buff: be used as the interior poke of buffer memory, unit: KB

Swap

Si: exchange to the exchange number of pages amount of internal memory, unit: KB/ second from disk

So: exchange to the exchange number of pages amount of disk, unit: KB/ second from internal memory

IO

Bi: send to the piece number of block device, unit: piece/second

Bo: from the piece number that block device receives, unit: piece/second

Analyze:

1) if r much larger than the number of cpu, system will be slack-off, if r greater than 4 times of the cpu number, illustrative system is faced with the cpu shortage of power, system can be seriously slack-off

2) if the id value often is 0, and sy+us is often greater than 80%, and illustrative system cpu overload lacks the cpu resource in other words.

3) if the wa value, illustrates disk greater than 25% not by good balance, system waits for magnetic disc i/o the cost plenty of time.Perhaps the load of disk is very heavy.

4) linux at first uses physical memory, re-uses the swap subregion after physical memory uses, thus linux system general memory utilization factor all than higher, if but si, so equals 0, illustrates that the swap subregion is not used.

Improvement project:

1) adjusts application, make its better utilization cpu, increase cpu number or rising cpu dominant frequency

2) optimize disk, solve the I/O problem

Iostat analyzes

Key parameter:

RkB/s: per second is read the K byte number.Be half of rsect/s, because every sector-size is 512 bytes.

WkB/s: per second is write the K byte number.Be half of wsect/s.

Avgqu-sz: average I/O queue, IOQ length.

Await: the stand-by period (millisecond) of average each equipment I/O operation.

Svctm: the service time (millisecond) of average each equipment I/O operation.

%util: had percent what time to be used for the I/O operation in one second, have how long I/O queue, IOQ is a non-NULL in one second in other words.

Analyze:

1) if disk shows high for a long time read-write, svctm is also much larger than 30 simultaneously, and the value of await is also much larger than svctm, and there is serious magnetic disc i/o bottleneck in illustrative system so.

2) if %util always near 100%, the I/O request that illustrative system produces is too many, may there be disk bottleneck in system

3) size of wait generally depends on the length of service time (svctm) and I/O queue, IOQ and the pattern of sending of I/O request.If svctm, illustrates that I/O does not almost have the stand-by period relatively near await; If await, illustrates that I/O queue, IOQ is oversize much larger than svctm, it is slack-off to use the response time that obtains.

Improvement project:

1) adjusts application, make it use disk i/o more efficient, can pass through the cache that revises disk queue, uses application server.

2) file system is distributed on 2 or a plurality of disk, changes disk raid scheme.

3) change better dish battle array or disk, use storage scheme faster.

The windows platform

Windows platform test, we will get object at the selected performance monitoring in performance monitor the inside, are being each object interpolation performance monitoring counter then, the performance monitoring object that mainly comprises, and the counter that comprises of each object:

1) processor processor

Processor performance object comprises the counter of weighing processor activity aspect.Processor be computing machine count and logical calculated, the associate member startup operation and the operation processing threads part.A computing machine can have many processors.Handler object is with the example of every processor as object.

Mainly comprise counter:

%Processor Time-finger processor is used for carrying out the number percent of non-idle thread time.Computing method are the time of the at interval interior non-idle thread activity of measurement example, to deduct this value at interval with example.(every processor has an idle thread, and this thread is cycles consumed when not having other threads to move).This counter is the main explanation device of processor activity, is presented at example observed rush hour average percent at interval the time.This value deducts this with 100% and serves inactive Time Calculation and come out.

2) memory internal memory

Memory performance object is made up of the physics on the description computing machine and the counter of behavioral virtual memory.Physical memory refers to the quantity of the random access memory on the computing machine.Virtual memory is made up of the space on physical memory and the disk.Many internal memory counters monitor pagings (refer to disk and physical memory code and data page mobile).Too much paging (a kind of performance of low memory) can cause delay, can influence the total system treatment effeciency.

The counter that mainly comprises:

Available MBytes is meant the physical memory amount of representing with MB, and this internal memory can distribute to a process at once or system uses.It equals to distribute to the summation of standby (buffer memory), free time and zero page or leaf list memory.Obtain the detailed explanation of memory manager, see also system performance and key to difficulty guide chapters and sections in MSDN and/or the Windows Server 2003Resource Kit.

3) Physical Disk hard disk

Physical Disk performance object comprises the hard disk on the supervisory computer or the counter of fixed disk drive.Disk is used for storage file, program and paged data and by reading these projects of retrieval and writing by record it is changed.The value of physical disk counter is the summation of logic magnetic disc (being divided into by disk) value.

The counter that mainly comprises:

Avg.Disk Queue Length refers to read and write the average of request (being that selected disk is lined up) in the example interval.

%Disk Time refers to that selected disc driver is busy with providing the service number percent of used time for reading or writing request.

4) Network Interface network

Network Interface performance object comprises the counter of weighing by the speed of a TCP/IP network connection transmission and reception byte and packet.It comprises the counter of supervising connection error.

The counter that mainly comprises:

Bytes Total/sec is the speed that sends and receive byte on each network adapter, comprises the frame character.NetworkInterface Bytes Received/sec be Network Interface BytesReceived/sec and Network Interface the summation of Bytes Sent/sec.

6. appendix

For the test data that obtains, may need following processing, use utraledit:

Regular expression:

Replace indentation: %[^t]+replacing with: # does not fill in any character or space

Replace blank line: %[^t] ++ ^p.

Claims (1)

1. the method for performance data of server collection and analysis, it is characterized in that, with data acquisition and analysis is core, by arrangement and the standard to acquisition method and target, forms normalizing operation, the critical data that can reflect server performance is gathered, with data is foundation, each subsystem of server is analyzed, thereby found the server performance bottleneck, the lifting server performance provides foundation in test or the practical application in order to solve, and concrete steps are as follows:

One, data acquisition scope

1) in test process, write down the test data in the whole test process,

2) determine the object of performance monitoring, comprise each subsystem, mainly comprise cpu, internal memory, disk, network;

Two, data acquisition synchronism

1) test macro may be divided into multi-layer framework, and the data aggregation on each layer architecture will guarantee temporal synchronous, could make accurately the state variation of system's run duration like this and describing;

2) the performance monitoring data of each subsystem will go up synchronously the assurance time;

3) test log: in test process, all will finish test log every day, writes down one day test process, goes wrong;

Three, collecting method:

The collection of data according to the difference of platform, is divided into two classes: Linux and windows, wherein:

Linux system comprises

1) performance monitoring instrument

A) cpu, internal memory: vmstat

B) magnetic disc i/o: iostat

C) network: sar

The Windows system

2) performance monitoring instrument

In the following order, open performance monitor: control panel-management tool-performance

Four, test data arrangement

The arrangement of test data, the test data formization is perhaps graphical, for convenient follow-up analysis is prepared;

1) test data patternization

At linux system, the data that we collect under the linux system all are text formattings, need us these graphical datas, are convenient to follow-up data analysis contrast;

Figure comprises broken line graph or histogram;

2) file designation rule

Naming rule:

A) allow filename comprise more fileinfo as far as possible;

B) form: instrument abbreviation _ platform information _ test model;

Five, method of testing:

1) sampling time interval: in once testing together, the sampling time interval of different instruments need equate;

2) adopt point value: sampled point is a numerical value of N,

A) test duration is below 1 hour: 150＜N＜250;

B) test duration is 1-2 hour: 250＜N＜450;

3) use of vmstat: at first need to create directory, be used for depositing test data:

[root@nodel~]#?cd/tmp/

[root@nodel?tmp]#?mkdir?test_vmstat

Carry out as issue orders: [root@nodel tmp] #vmstat5n 〉/tmp/test_vmstat/file_name# per 5 second value once, and the result outputed in the file;

4) use of iostat: at first need to create directory, be used for depositing test data: order as follows:

[root@nodel~]#?cd/tmp/

[root@nodel?tmp]#?mkdir?test_iostat

Carry out as issuing orders:

[root@nodel tmp] # iostat-x5n|grep sdn 〉/tmp/test_iostat/file_name# per 5 second value once, and the result outputed in the file;

5) use of sar: at first need to create directory, be used for depositing test data: order as follows:

[root@nodel~]#?cd/tmp/

[root@nodel?tmp]#?mkdir?test_sar

Carry out as issuing orders:

[root@nodel tmp] # sar-n DEV5n|grep eth0 〉/tmp/test_sar/file_name# per 5 second value once, and the result outputed in the file;

Six, windows performance monitoring instrument: control panel-management tool-performance, use step:

1) counter daily record: open the performance monitoring instrument, click performance daily record and alarm, this step is finished in the gated counter daily record, and the syslog file that an acquiescence is arranged is seen in the perform region on the right: system overview

2) newly-built daily record setting: the perform region on the right side, click by right key, select newly-built daily record setting, eject dialog box, insert Log Names;

3) add counter: click and add the counter button, eject dialog box, in the performance object the inside, the object that selection will be monitored comprises processor, internal memory, network;

The corresponding some counters of each object are chosen gated counter from tabulation, after adding the counter that needs and finishing, carry out next step

4) journal file is set: above step is provided with for conventional, and the click logs file carries out the setting of journal file now; Keep acquiescence herein;

5) plan is set: be set to manually carry out or as required, plan to be provided with;

6) test beginning starts, and Icon Color has redness to become green;

Newly-built daily record is finished, and just sees newly-built daily record in counter daily record workspace; Red icon is a halted state, and green is a running status; Choose daily record, click right or manual the execution begin and stop;

Six test datas are understood and are analyzed

About the Linux platform: in the test process, by vmstat, iostat, sar surveillance resource mainly comprises: process, internal memory, swap, cpu, magnetic disc i/o and network interface card;

Vmstat analyzes: key parameter:

1) process: we mainly pay close attention to r and b parameter:

The r parameter: the operation queue number of threads, comprise the thread that moving and the thread of waiting for CPU timeslice, this numeral has at least a thread to want waiting for CPU greater than the number of CPU, and the thread of waiting for CPU is many more, might exert an influence to performance more;

The b parameter: the kernel thread average in the per second waiting list, just comprise thread at the I/O of pending file system, or owing to the internal memory thread that control is suspended of packing into;

Process is owing to the internal memory control of packing into is suspended, and the obstruction row (b) in vmstat report show that number of threads increases, rather than number of threads increases in the operation queue;

2) cpu: use the number percent segmentation of CPU time in this time interval, the cpu row are as follows:

Us parameter: us row have shown the CPU time that user model consumed;

Sy parameter: sy row have shown that in detail CPU carries out the number percent that thread is taken time under system model;

Id parameter: id row have shown the number percent of CPU free time when not having unsettled magnetic disc i/o or stand-by period, do not have unsettled I/O, and the time of the wait that is useful on was included in free time;

Wa parameter: wa row have shown the percentage of time of CPU free time when unsettled magnetic disc i/o is arranged in detail, has a uncompleted magnetic disc i/o when moving at least when waiting for, this time is included into the time of waiting for I/O, this process is used asynchronous I/O, then the I/O to disk asks invokes thread to be got clogged or sleep, and is done up to request; In case the I/O of process request is finished, this process is placed in the operation queue, and I/O finishes very soon, and this process is used more CPU time; The value of wa should not surpass 25%;

3) other parameters

Memory

Swpd: virtual memory operating position, unit: KB;

Free: idle internal memory, unit K B

Buff: be used as the interior poke of buffer memory, unit: KB;

Swap

Si: exchange to the exchange number of pages amount of internal memory, unit: KB/ second from disk;

So: exchange to the exchange number of pages amount of disk, unit: KB/ second from internal memory;

IO

Bi: send to the piece number of block device, unit: piece/second;

Bo: from the piece number that block device receives, unit: piece/second;

Analyze:

1) r is much larger than the number of cpu, and system will be slack-off, and r is greater than 4 times of the cpu number, and illustrative system is faced with the cpu shortage of power, and system can be seriously slack-off;

2) the id value often is 0, and sy+us is often greater than 80%, and illustrative system cpu overload lacks the cpu resource in other words;

3) the wa value illustrates disk not by good balance greater than 25%, and system waits for that the load of magnetic disc i/o or disk is very heavy the cost plenty of time;

4) linux at first uses physical memory, re-uses the swap subregion after physical memory uses, thus linux system general memory utilization factor all than higher, si, so equal 0, illustrate that the swap subregion is not used;

Improvement project:

1) adjusts application, make its better utilization cpu, increase cpu number or rising cpu dominant frequency;

2) optimize disk, solve the I/O problem;

Iostat analyzes

Key parameter:

RkB/s: per second is read the K byte number, is half of rsect/s, because every sector-size is 512 bytes;

WkB/s: per second is write the K byte number, is half of wsect/s;

Avgqu-sz: average I/O queue, IOQ length;

Await: the stand-by period (millisecond) of average each equipment I/O operation;

Svctm: the service time (millisecond) of average each equipment I/O operation;

%util: had percent what time to be used for the I/O operation in one second, have how long I/O queue, IOQ is a non-NULL in one second in other words;

Analyze:

1) disk shows high for a long time read-write, and svctm is also much larger than 30 simultaneously, and the value of await is also much larger than svctm, and there is serious magnetic disc i/o bottleneck in illustrative system so;

2) %util is always near 100%, and the I/O request that illustrative system produces is too many, and may there be disk bottleneck in system;

3) size of wait generally depends on svctm and the pattern of sending of the length of I/O queue, IOQ and I/O request service time, and svctm illustrates that relatively near await I/O does not almost have the stand-by period; Await illustrates that much larger than svctm I/O queue, IOQ is oversize, and it is slack-off to use the response time that obtains

Improvement project:

1) adjusts application, make it use disk i/o more efficient, by revising the cache of disk queue, use application server;

2) file system is distributed on 2 or a plurality of disk, changes disk raid scheme;

3) change better dish battle array or disk, use storage scheme faster;

About the windows platform

The windows platform test will be at the selected performance monitoring object in performance monitor the inside, and for each monitored object adds the performance monitoring counter, the counter that performance monitoring object and each object comprise is as follows then:

1) processor processor

Processor performance object comprises the counter of weighing processor activity aspect, processor be computing machine count and logical calculated, the associate member startup operation and the operation processing threads part, a computing machine has many processors, and handler object is with the example of every processor as object;

Mainly comprise counter:

%Processor Time-finger processor is used for carrying out the number percent of non-idle thread time, computing method are, measure the time of the at interval interior non-idle thread activity of example, deduct this value at interval with example, this value is that every processor has an idle thread, cycles consumed when this thread moves when not having other threads, this counter is the main explanation device of processor activity, be presented at example observed rush hour average percent at interval the time, this value deducts this with 100% and serves inactive Time Calculation and come out;

2) memory internal memory: Memory performance object is made up of the physics on the description computing machine and the counter of behavioral virtual memory, physical memory refers to the quantity of the random access memory on the computing machine, virtual memory is made up of the space on physical memory and the disk, many internal memory counters monitor paging, too much paging causes delay, can influence the total system treatment effeciency;

The counter that mainly comprises has: Available MBytes is meant the physical memory amount of representing with MB, and this internal memory can distribute to a process at once or system uses, and it equals to distribute to the summation of standby, free time and zero page or leaf list memory;

3) Physical Disk hard disk: Physical Disk performance object comprises the hard disk on the supervisory computer or the counter of fixed disk drive, disk is used for storage file, program and paged data and by reading these projects of retrieval and writing by record it is changed, and the value of physical disk counter is the summation of logic magnetic disc value;

The counter that mainly comprises has: Avg.Disk Queue Length refers to read and write the average of request;

%Disk Time refers to that selected disc driver is busy with providing the service number percent of used time for reading or writing request;

4) Network Interface network: Network Interface performance object comprises the counter of weighing by the speed of a TCP/IP network connection transmission and reception byte and packet, and it comprises the counter of supervising connection error;

The counter that mainly comprises has:

Bytes Total/sec is the speed that sends and receive byte on each network adapter, comprise the frame character, Network Interface Bytes Received/sec be Network Interface BytesReceived/sec and Network Interface the summation of Bytes Sent/sec;