Re: Implementing hrProcessorLoad

[[e-mail]]

Hi George,

>     
>     If any of the RFC 1514 authors or contributors are listening, could 
>     you please describe the implementation approaches that came to mind 
>     when defining this object?
>     

A cheap and dirty way (at least on SunOS4, SunOS5, HP A.09.x, and 
AIX 3.x) is to use a system utility like `iostat' or `vmstat' and 
parse the output. Both `iostat' and `vmstat' support an interval 
option. To get the average CPU usage over the last minute, just
use a pipe and invoke the utility with an interval of 60 seconds.
You should get something back over the pipe every 60 seconds or so.
For sun4m machines I would probably use `mpstat' so that I would get 
the cpu usage for each cpu.

*NOTE* Accessing the kernal and the /dev/kmem file directly offers 
       better performance, given that there is less overhead.
       The iostat/vmstat will occur the cost of two additional processes
       (e.g., one for the iostat/vmstat call and one for the shell)
       plus context switches.  It can be worst if you use awk/grep/sed
       to parse the output.
       
>
>     Has anyone implemented this object on a UNIX platform without having  
>     to wait an entire minute (blocked) while the average is computed?
>     

This would not be a problem if you added the pipe to the same `select' call
you are using for the incoming SNMP port.

By the way, 

marktom

-------
Mark Tom <[e-mail]>
NetLabs, Incorporated
4920 El Camino Real
Los Altos, CA 94022
[phone] (work)/[phone] (fax)