I am developing a server program that send 3 types of data
I data - 13888 usecs
P data - 2777 usecs
B data - 555 usecs

1. This data are to send within the time frame given.

2. There is a counter to add up the ideal time the data should be send known as streamtime.

3. At the beginning of the program, i used gettimeofday to get the time known as start.

4.Before i send the data, i gettime again known as beforesend

5. i get the difference between beforesend and start and then compare with streamtime. If is more than streamtime it wont send.

6. After sending i gettime again , i used the time got and minus it from streamtime, this will be the amount of time i will sleep.

7. This will continue in a loop until the program terminates.

My problem is that , sometimes at 5. the difference which is the timing is more than streamtime. I send 1700 pieces of data , about 400 pieces of data have that time more than the stream time.

Is is because of the background process tat cause some of this time to increase a lot?

Sort of. You are running in a system that gives out cpu time in slices (quanta). If some process that has nothing to do with what you are doing needs time, it will get it.

You need to develop some kind of inter-thread communication scheme using sempahores or whatever is appropriate. Just guessing that streamtime will be 13000 ms and then sleeping that long will get you in trouble. You need to have each transmission thread signal <I am Done>
and then have one controlling thread wait until it gets all three of these signals or semaphores or whatever, before it tries to send the next batch of data.

dont really understand , i quite new to linux programming
read up it should work by using sched_setscheduler, mlockall and gettimeofday

but i dont know how to implement it

Well, because of the nature of linux, or windows, or any non-realtime OS you can't guarentee timing.

but i read up , they say non real time OS still could implement it?

You can get relatively close, but most of the time you'll find people implement protocols for what data should look like (start sequence, stop sequence, check sum) rather then timing schemes for how data should come because they are safer. Technically speaking, if the kernel decides it needs to do something more important it could cut you off in the middle of your routine. You could check out the RTLinux extensions and implement a kernel module for the timing that would easily get you the resolution you want.

One of the timing issues I imagine you'll run into is the fact that it is very hard to syncronize the clock on two different machines across the network, especially for us resolutions. And if you get off sync you'll end up with a huge mess.

hmm if i use rtlinux , my problem will be automatically slow or do i have to do something to make it work real time?