Go Nmon server monitoring, indicator description

First, the Nmon in the parameters of the indicators:

Sys_summ: Shows the overall performance of the current server

Total System i/ostatistics:
Avg TPS during an interval: Displays the average number of disk I/OS during the acquisition interval, equal to the average of io/sec columns in the sheet disk_summ.
Max TPs During an interval: Displays the maximum number of disk I/Os in the acquisition interval, which is equal to the maximum value of the Io/sec column in sheet disk_summ.
Max TPs Interval Time: Displays the disk maximum I/O at the same point.
Total number of Mbytes read: Displays the overall megabytes of disk reads during the acquisition interval, which may be a nmon bug, is not accurate, and virtual disks partitioned using LVM may have duplicate statistics.
Total number of Mbytes written: shows the amount of megabytes written in the disk during the acquisition interval, which is not accurate for the same reason.
Read/write Ratio: Displays the value of total number of Mbytes read/total the Mbytes written. Real-time Read and write ratios
Io/sec: Shows only the graph of the disk io/sec, excluding network I/O. Number of transfers per second to physical disk

Such as:

Cpu:
users%: Displays the time ratio (AVG, Max) of all CPUs in user mode during the acquisition interval.
sys%: Displays the time ratio (AVG, Max) of all CPUs in system mode during the acquisition interval.
wait%: Displays the percentage of time that all CPUs in the acquisition interval are idle and waiting for I/O to complete (wait% is a CPU idle state, and when the CPU is idle and the process is in the D State (non-disruptive sleep), the system counts the time and calculates it to wait%), and wait % is not a time value, but a ratio of time, so the more server CPU, the lower the wait%, in the same I/O Wait time, it represents the ratio between I/O operations and computational operations. For I/O intensive applications generally wait% high, and sheet proc blocked is also high, this time need to pay attention to what led to excessive process waiting.
Idle%: Displays the percentage of idle time for all CPUs in the acquisition interval (AVG, Max).
CPU%: Displays the user%+system% of all CPUs during the acquisition interval.
2. AAA
Displays current server basic information, such as operating system version, current LPAR name, acquisition time and number of times, etc.

3. Straylines
Displays the collection values that were not generated in this Nmon analysis file.
4. BBBP
As the contents of this sheet are many, see (partial):

A
For example, displays the underlying resource information for the current server, and the current server operating system is the Cent OS version.


Memtotal: Displays the current server physical memory size, this server has 8063180 kb≈7874 MB or so.
Memfree: Displays the current server's free memory size, this server has 5052336 kb≈4934 MB or so.
Buffers: Display the current server buffer (in memory to write to disk) the size of the cache, the server has 459108 kb≈448 MB or so, note that the value here is only the initial acquisition of static values, the specific buffer changes also need to see sheet MEM.
Cached: Displays the current server cache cache size (read from disk to memory), the server has 1032572 kb≈1008 MB or so. , the value here is only the static value at the beginning of the acquisition, the change of the specific buffer also need to see sheet MEM.
Swapcached: Displays the size of the current server swap space that has been cached, and the server has not been used to swap space.
Swaptotal: Displays the current server swap space size, this server has 8385532 kb≈8189 MB or so.
Swapfree: Displays the current server swap free space size, the server swap space is idle.

Because the system group permissions are different when performing nmon, the information in the BBBP disk may be missing, such as the disk information that is displayed after the root permission is executed Nmon the generated file, and you can see the size of each disk and the partition usage under disk.
5. Cpu_all
Displays the utilization of all CPUs in the current server during the acquisition period, displayed by time and user%, system%, wait%.

The current server has a total of 4 CPUs (CORE) 8 cores.
In general, the CPU utilization in the user% should account for about 70%, sys% should account for about 30%, if sys% or wait% accounted for or over the user% should pay attention to what caused excessive system consumption, may be a large number of disk or network I/O.
For example, this project with the increase in concurrency, the application process to the CPU consumption increased on the wait%, the investigation is due to NFS read and write encountered bottlenecks caused by:

6. Cpu_summ
Displays the utilization of all CPUs on the current server, with 4 CPUs (Core) on the current server, each with a different CPU load.

7. Disk_summ
Displays the rate of read/write for all disks and partitions (KB/S) and the I/O rate for all disks and partitions by acquisition time.The io/sec of a acquisition time point equals the sum of io/sec of all disks and partitions at that point in the sheet diskxfer. Therefore, the I/O values are duplicated at this point in time! In addition, I/O in this sheet does not include I/O in NFS.





As Wavg by Nmon Guide is to remove the value of the 0 value in order to close to the true average, but the formula of Wavg (the sum of all the values in the computed column, divided by the sum of all the values in the column) is not simply to remove the 0 value, It can be understood here that Wavg is closer to the average of resource consumption than AVG, so it is recommended to focus on WAVG in all resources sheet.
IBM redpaper, "Linux performance and Tuning guidelines," describes Linux's I/O subsystem architecture as follows:

Most of the Nmon (including iostat) of system I/O are from/proc/diskstats, and these values come from the block layer, and the logical volume in LVM will "visible as a standard block Device ", so real disks, LVM logical volumes, partitions, and so on are all shown here, and will be counted when Nmon calculates the total value.
Disk Read/write KB is the sum of the values of the row (all disks and partitions) in the sheet diskread and diskwrite at the same acquisition time point, which inevitably includes duplicate values, such as a time when the SDA disk has been Write-KB, where SDA1 partition write Kb,sda3 partition write KB, this time disk write should be a KB, but here will be repeated statistics partition value, resulting in disk write KB. Disk I/O also has the same problem!
It is also important to note that in some nmon generated files, the title metric in the figure is KB (lowercase)/s, but the actual statistic is KB (uppercase)/S.

http://blog.csdn.net/he_jian1/article/details/41039709/

Go Nmon server monitoring, indicator description