High run queue from vmstat 1

[gstrever]

I have the following vmstat sample. Can anybody tell me if this should be something to worry about or is it ok. The system is a 16 way S85.
The values of the fields not displayed are all 0.

kthr memory faults cpu
----- ----------- ----------- -----------
r b avm fre in sy cs us sy id wa
18 5 2072960 680850 8724 51989 30109 70 17 11 1
18 4 2072960 680696 8887 52089 29443 72 16 11 0
16 4 2072960 680571 7321 49911 30490 69 17 13 1
20 5 2072924 680417 8310 73632 30070 70 19 11 1
15 5 2072989 680204 8396 70243 29539 70 18 11 0
23 4 2072844 680255 8090 57420 25552 79 16 5 0
17 5 2073224 679639 8089 79716 28776 74 18 7 1
22 4 2072593 680112 8133 88704 26993 72 19 8 1
23 4 2072587 680044 7862 189142 29758 66 22 12 1
11 5 2072890 679547 9452 63938 36455 62 21 16 1
17 4 2073044 679231 8737 62049 37173 59 22 18 1

What is the "thumbsuck" value for the run queue for a 16 way server and does the idle % mean all is still well ?

 

[stawolg]

In the IBM literature (RS6000 Performance tools in focus)
there are given 3 rules of thumb to estimate the load of a machine:
1) The number of the runnable threads (r) should be less than 5 times the number of unbound processors, so your are far away from this limit. In your example, on average there is approx 1 thread per processor, and thats nice
The r value is nearly equal to the load value output of the "w" or "uptime" command.
2) The number of blocked thread (b) (a thread waiting for e.g. i/o) should be less than 3 on a single proc system. As your wait for i/o is neclectable, it seems ok.
3) usr+sys should be less than 80% to have some remaining ressources.

Whats the conclusion? You have some nice computational load on your machine, but if it is too much or just ok can only be judged by the user who is waiting for his processes to be finished!