Identifying CPU Bottlenecks with vmstat kthr / r

Question

Hi,
Can someone help me with the interpretation of vmstat with CPU and multiples cores, threads ?
In short, the server is experiencing a CPU bottleneck when “r” is greater than the number of CPU’s on the server.
But what should I use in these cases for example :
2 CPU's dual cores should mean 4 CPU's ?
2 CPU's dual cores dual threads should mean 4 CPU's or 8 CPU's ?
# vmstat 5
     kthr   memory          page             disk      faults        cpu
     r b w swap  free re mf pi p fr de sr s0 s1 s2 s3  in  sy  cs us sy id
     12 0 0 11456 4120 1  41 19 1  3  0  2  0  4  0  0  48 112 130  4 14 82
     14 0 1 10132 4280 0   4 44 0  0  0  0  0 23  0  0 211 230 144  3 35 62
     15 0 1 10132 4616 0   0 20 0  0  0  0  0 19  0  0 150 172 146  3 33 64
     17 0 1 10132 5292 0   0  9 0  0  0  0  0 21  0  0 165 105 130  1 21 78Thanks !

Answer 1

Look at LAT from prstat (generally I use prstat -amL). This will give you an idea of the delay due to wait for CPU.
Looking at your vmstat:
# vmstat 5
     kthr   memory          page             disk      faults        cpu
     r b w swap  free re mf pi p fr de sr s0 s1 s2 s3  in  sy  cs us sy id
     12 0 0 11456 4120 1  41 19 1  3  0  2  0  4  0  0  48 112 130  4 14 82
     14 0 1 10132 4280 0   4 44 0  0  0  0  0 23  0  0 211 230 144  3 35 62
     15 0 1 10132 4616 0   0 20 0  0  0  0  0 19  0  0 150 172 146  3 33 64
     17 0 1 10132 5292 0   0  9 0  0  0  0  0 21  0  0 165 105 130  1 21 78Note that there is a run queue but also much cpu idle - 62% to 82%
Use mpstat to see the load on each cpu.
If you feel that there is a performance problem with the application(s) running on this server I don't think this vmstat supports the need for more CPUs.
What is the hardware and Solaris level involved.
have a good day,
Glen
(PS I've been wrong before so don't rule out a need for more CPUs - but collect more data before reaching a conclusion.)