AIX performance tuning-memory and CPU

AIX Performance Tuning-Memory and CPU

Sar-P all CPU usage

Sar-A File Access
The number of fast reads (B) by the Directory Access daemon when the dirblk/s locates the file
Iget/s I node to find the number of times the system process is called
Lookuppn/s Directory: the number of times the process finds the V node and obtains the path name.

Activity of SAR-B buffer, including transmission, access, and hit rate
Number of bread/s and bwrit/s block Io operations
Number of lread/s and lwrit/s logical IO requests
Number of I/O operations on pread/s and pwrit/s bare Devices
% Rcache, % rwrit cache hit rate. formula: (lreads-breads)/lreads) * 100

Sar-C system calls
Total number of exec/s, fork/s calls, and execution system calls
Number of sread/s, swrit/s Read/writ system calls
Number of characters that rchar/s and wchar/s are called by the Read/Writ System
Total number of scall/S system calls

Sar-K kernel process activity
Number of kernel processes interrupted by kexit/s
Kproc-OV/s cannot create kernel processes due to the number of processes
Number of kernel processes assigned by jobs in ksched/s

Sar-m message queue and Traffic Signal Activity
MSG/s IPC Message Queue Activity
SEMA/s traffic signal Activity

 

Sar-D disk read/write status

Sar-Q queue statistics
Average number of Run-SZ kernel threads in the running queue
% Runocc running queue usage in the recent time period
Average number of swpq-SZ kernel threads waiting for page Scheduling
% Swpocc swap queue recent activities

Sar-r page scheduling information
Number of page replacements per second in cycle/s
Number of page fault times per second in fault/s
Number of free pages of slots in the page Space
ODIO/s no page space disk I/O count per second

Sar-V process, kernel thread, I node, and file table status

Sar-W context switch count

Sar-y tty device activity
Canch/s tty Input QueueSpecificationsNumber of characters
Mdmin/s tty modem interrupted
Outch/s output queue characters
Rawch/s input queue characters
Revin/s tty receiving interrupted
Xmtin/s tty Transmission Interrupted

 

If the CPU usage is close to 100% (USR + system), it can be considered as a CPU bottleneck. If a considerable amount of time is spent on Io wait, it means that CPU execution is limited by disk Io,
The IO bottleneck may be caused by file access or insufficient memory to allocate pages.
Note: the time spent by the system waiting for remote file access is not recorded in the I/O wait time. If the CPU and I/O wait time are both quite low, but the response time is not satisfactory, check that the system
The amount of time it takes to wait for remote I/O. As a result, there is no command in Aix to analyze remote I/O, and you can only observe it by tracking data.

Vmstat

The vmstat Command reports kernel threads, virtual memory, disks, traps, and CPU activity.
Kthr thread Activity
R running queue
B waiting queue

Memory virtual and actual memory usage
Virtual page of AVM Activity
Fre idle page. When the system memory is larger than 64 MB, the minimum minfree value is 120 frames. When the memory is smaller than 64 MB, the minimum value is twice the memory size in MB.
Minfree and maxfree values can be viewed using the vmtune command.

Page page fault and page activity. When a page (non-NFS or permanent file page) is allocated in memoryWorkPage, the work page usually includes the application stack,
Data and other shared memory segments. Therefore, when a program stack or data area needs to grow, the memory will be accessed, and VVM will allocate space from the devices where Ram and page space are located. This means
The page space is used before the memory is exhausted.
Re page Input and Output list, the amount of memory recycled per second, when the page is in the idle list and not reused, it will be recycled should be initialized without a new Io, these include incomplete I/O operations that are used by vmm.
Pre-read the page on which the algorithm is transferred to the memory.
Pi from the page space page in page
Po page from page space out

Fr page idle (page reset)
Number of times the SR page is scanned by the page Scheduling Algorithm
Cy page Scheduling Algorithm for scheduling clock cycle

Faults trap and system interruption rate
In device interruption
Sy system call
CS kernel thread context switch

CPU usage
USR user process
Sys system process
Id CPU idle time
Wa wait disk Io time

Principles:
R <5, B ≈ 0,
If FRE is less than minfree, continuous page scheduling will occur, leading to system performance problems.
For page columns, re, PI, Po, and Cy, the PI rate is not more than 5. If pagin occurs, the associated page must be pageout first.
In an environment with relatively insufficient memory, pagein forces steal operations on different pages. If the system is reading a large number of permanent pages, you may see the Po and PI Columns
Inconsistent growth may occur, which does not necessarily indicate that the system load is too heavy, but it is necessary to check the data access mode of the application. In a stable situation, the scanning rate and Reset rate are almost the same
When multiple processes use different pages, the pages are more unstable and messy, and the scanning rate may be higher than the Reset rate.

The faults column, in, Sy, and CS will continue to jump. There is no definite limit here. The only difference is that these values must be at least 100.

The CPU columns, US, sys, ID, and Wa are also uncertain. The ideal status is to make the CPU in the 100% running state. This is only suitable for single users.
If the "us + sys" 80 in a multi-user environment is used, the process will wait for time in the running queue, and the response time and throughput will decrease. Wa> 40 indicates that there may not be an unreasonable balance between disk Io, or disk operations are frequent,

Tag:Performance Tuning