Linux Performance monitoring
Performance monitoring is a hard job for system admin. It depends
on the application running on the box and also the experience of the sys admin.
Here are some basic guide for performance measuring.
|
|
Good
|
warning
|
Bad
|
|
CPU
|
user+sys < 70%
|
70%<user+sys<90%
|
user+sys > 70%
|
|
Memory
|
no swap
|
per CPU < 10 pages/s
|
lots of SI/SO
|
|
DISK
|
iowait < 20%
|
20% < iowait < 50%
|
iowait > 50%
|
Here we will show about the performance measuring tools and
the output.
1.
vmstat:
commands: vmstat [interval] [counters]
vmstat
is a useful tool to analyse the performance.
Procs: process number:
r:
runnable process but waiting for CPU resource. If there are lots of r, it
usually means the cpu resource is not enough (experience value < 2/cpu)
b:
blocked processes waiting for other resource (usually memory or IO)
Memory: memory information:
swpd:
swaped memory measured by kb
free: usable memory
buff: buffered memory
cache:
cached memory
swap: swap information
si:
swap in. from disk to memory
so:
swap out. From memory to disk
IO: block IO information:
Bi:
block in. read from disk to memory
Bo:
block out. Write the memory information to disk.
System:
In:
interruption numbers
cs:
content switch numbers
there
shouldn’t be many Interruptions and content switching for a stable system.
CPU: cpu usage by percentage
Us:
user level, sy: system level, id: idle. Wa: wait for IO.
free command:
Free command is very simple. Just need to
understand the
Buffers: blocked IO related memory
Cached: file content cached.
Uptime:
Showing the current time, system running time since the last reboot and
the current users. It will show the system load for the last min, 5 mins and 15
mins. It should consider the cpu cores when measuring the load.
iostat: check the disk IO status
The most important value is the %iowait: it
is usually beblow 20%
%util. the lower the better.
No comments:
Post a Comment