Hi,

I'm creating a multi-agent program, in which I have to know the exact
cpu usage of my child processes.

in windows there is a function named "getProcessTimes(int pid);",
is there any appropriate linux alternative function that returns the amount
of instructions executed by a process or amount of time used by the child
process, from the process birth until now.

thank you in advance for your help.

First, keep track of all your children's process ID's by noting what the fork() function returns. Then, for each process ID (567, say), read the file [/B]/proc/567/stat[/B]. Two of the numbers returned are exactly what you want. For information on the numbers you get back, do this at the command prompt:

Hope this helps.

Hi,
Thank you for reply,
but the thing is, the utime, and stime,
do not change very often. These two change every 2 3 seconds
and n my experiment they increment one step each time.
Also is there any faster way to avoid reading the file each
time I want to know the process times?

Thank you.

I'm not aware of any better solution. Sorry.

Alright , so you need to figure out how much cpu power each of ur child proc's is consuming , hmmm.
not that I'm sure about that but I guess you could eventually calculate that yourself by measuring how much time each process has been taking (between two points in your code , or so)

I'm going to assume that each of your processes runs an infinite loop of some sorts , then you could insert cputick counting functions at the start and end of the iteration (a time counter if ya know what I mean) , then when you know how much time each iteration of each process has taken you probably can find out the cpu usage of each one by dividing the overall cpu usage by the process specific time consumption...etc

Hope that gets you somewhere,

Regards

It may help if you re-state the problem, so that you can come up with a metric that better answers your question or better assists you in making your decision, whatever it is.

"The CPU load of a process" is actually a fairly abstract number: it depends upon the nature of the machine, the nature of the workload presented by this application, and the "ambient load," which is the workload presented by all the other processes that are running at the same time. As a result, therefore, it's probably not too useful.

A much better metric would be some sort of metering or counter that the process generates for itself, and which is somehow meaningfully related to the work that this process is designed to do. For example, the number of requests completed per-second, or the ratio of the total application workload that was completed by this particular process. A still better metric might consider the totals of the various kinds of wait-time that are being experienced by the requests ... or by some random sample of those requests.