我是怎么分析nmon 的

我是怎么分析 nmon 的：

 

关键还是看参数

NMON---memuse选项
选项官方解释
Except for %comp, the values on this sheet are the same as would be reported by the vmtune command. 

%numperm the percentage of real memory allocated to file pages.
%minperm value specified on the vmtune command or system default of 20%. This will normally be constant for the run unless the vmtune or rmss commands are used during collection.
%maxperm value specified on the vmtune command or system default of 80%. This will normally be constant for the run unless the vmtune or rmss commands are used during collection.
minfree the minimum number of pages AIX is to keep on the free list.  Specified on the vmtune command or system default of maxfree - 8.
maxfree the maximum number of pages AIX is steal in order to replenish the free list.  Specified on the vmtune command or system default.
%comp The percentage of real memory allocated to computational pages. NMON_Analyser calculates this value.  Computational pages are those backed by page space and include working storage and program text segments.   They exclude data, executable and shared library files.

The Analyser generates two graphs.  The first shows the split between computational and file pages by time of day.  The second plots the values of %numperm, %minperm, %maxpermand %comp.

If %numperm falls below %minperm then computational pages will be stolen.   If %numperm rises above %maxperm then computational pages cannot be stolen.  Low values for both %minperm and %maxperm indicate that the system has been tuned for a database server.   You may also want to check the setting of STRICT_MAXPERM on the BBBP sheet (if present).

下面看一下客户的 nmon 分析报告：

 

2;
NMON—page选项
选项官方解释
This sheet has the paging statistics as recorded by NMON.

faults  the number of page faults per second. This is not a count of page faults that generate I/O, because some page faults can be resolved without I/O.
pgin  the total rate/sec of page-in operations to both paging space and file systems during the interval.
pgout  the total rate/sec of page-out operations to both paging space and file systems during the interval.
pgsin  the rate/sec of page-in operations from paging space during the interval.  This is the same as the pi value reported by vmstat. If pgsin is consistently higher than pgsout this may indicate thrashing.  
pgsout  the rate/sec of page-out operations to paging space during the interval. This is the same as the po value reported by vmstat.
reclaims from NMON 10 onwards this field is the same as thefrvalue reported byvmstatand represents the number of pages/sec freed by the page-replacement routine.
scans  the number of pages/sec examined by the page replacement routine.  This is the same as the sr value reported by vmstat. Page replacement is initiated when the number of free pages falls below minfree and stops when the number of free pages exceeds maxfree.
cycles  the number of times/sec the page replacement routine had to scan the entire Page Frame Table in order to replenish the free list.  This is the same as the cy value reported by vmstat but note that vmstat reports this number as an integer whereas nmon reports it as a real number.
fsin calculated by the Analyser as pgin-pgsin for graphing
fsout calculated by the Analyser as pgout-pgsout for graphing
sr/fr calculated by the Analyser as scans/reclaims for graphing

NMON_Analyser produces two graphs.   The first shows paging operations to/from paging space.  The ideal here would be no more than 5 operations/sec per page space (see the BBBC sheet for details).   The second graph shows the scan:free rate.   Memory may be over-committed when this figure is >4 although you also need to examine the MEM and PAGE sheets as well.

看一下 客户的nmon 报告：

 

 

NMON-topas选项
选项官方解释
 This sheet is only generated if you specify the -t  flag on the NMON command line.  The output is similar to that produced using the ps v command.   Note that, because of the limitation of having only 65,000 lines on a single sheet, some data may be omitted for very large files and this may mean that entire PIDs or even commands may be missing from the analysis.

Note that data are only present for processes that consumed a significant amount of CPU during an interval.  The TOP sheet does not represent a complete view of the system.

NMON_Analyser does the following:
? reorders the columns for easier processing.  
? Sorts the data on the sheet into COMMAND name order - using TIME as a minor sort key. 
? Creates a table at the end of the sheet summarising the data by command name and used for graphing.

You can see the detail section by scrolling to the top of the sheet.   The summary table is largely obscured by the graphs and so you will need to move (or delete) them for easier viewing.

PID  in the detail section this is the process ID of a specific invocation of a command. In the summary table this is the command name.
%CPU  in the detail section this is the utilisation of a single processor (rather than of the system) by that PID during the interval. In the summary table this is the average amount of CPU used by all invocations of the command during the collection period.
%Usr  in the detail section this is the average amount of User-mode CPU used by that PID during the interval.
%Sys  in the detail section this is the average amount of Kernel-mode CPU used by that PID during the interval.
Threads the number of (software) threads being used by this command.
Size  the average amount of paging space (in Kbytes) allocated for the data section (private segment + shared library data pages) for one invocation of this command.  This is the same as the SIZE figure on the ps v command.  Note that if Size is greater than ResData it means some working segment pages are currently paged out.
ResText the average amount of real memory (in Kbytes) used for the code segments of one invocation of this command.   Note that multiple concurrent invocations will normally share these pages.
ResData the average amount of real memory (in Kbytes) used for the data segments of one invocation of this command.  A method of calculating real memory usage for a command is ResText + (ResData * N).
CharIO  this is the count of bytes/sec being passed via the read and write system calls. The bulk of this is reading and writing to disks but also includes data to/from terminals, sockets and pipes. Use this to work out which processes are doing the I/O.
%RAM  this is an indication of what percentage of real memory this command is using. This is (ResText + ResData) / Real Mem; it is the same as the %MEM value on the
ps v command.    Due to rounding/truncation, and the large amounts of memory in modern systems, this is usually 0. 
Paging sum of all page faults for this process.  Use this to identify which process is causing paging but note that the figure includes asynchronous I/O and can be misleading.
Command name of the command
WLMClass name of the Workload Partition or Workload Manager superclass to which this command has been allocated (64-bit kernel only).  
IntervalCPU generated by the Analyser.   In the detail section this shows the total amount of CPU used by all invocations of a command in the time interval.  It is calculated as the sum of CPU used by all PIDS running the same command divided by the number of active processors (physical cores) available during the interval. In the summary section this is broken down as Average, Weighted Average and Maximum and is used to generate the graph.
WSet generated by the Analyser.   In the detail section this shows the total amount of memory used by all invocations of a command recorded in the time interval.  It is calculated as ResText + (ResData * N) (where “N” is the number of copies of this command running concurrently during the interval).  In the summary section this is broken down as Minimum, Average and Maximum and is used to generate the graph.
User generated by the Analyser if a UARG sheet is present.  This contains the name of the user running the process. 
Arg generated by the Analyser if a UARG sheet is present.  This contains the complete argument string entered for the command.

看一下客户的report: