This repo is about the first assigment on the real-time embedded systems class . We are called to modify a solution of producer consumer problem,which use threads , in order to be able to support multiple producer and multiple consumer threads, and having functions as queue items. The main programm of the assigment is prod-cons_multithreading.c
Statististis about the waiting time , are stored in stats folder
For executing the program you have first to compile it ,by executing the command bellow inside the project folder :
gcc prod-cons_multithreading.c myFunctions.c -pthread -o program.out -lm -O3
after compilation finish you are able to execute the programm with the following command
./program.out p q , where p is the number of producer threads you want to have and q the number of consumers,for instance if you want to have 4 producer threads and 4 consumers threads you have to run the command:
./program.out 4 4
If you want to run the programm for a compination of p and q values, you can edit the bench.sh file and set the p and q values , for which you want to run the programm, in the two for loops . After editing the file , by running :
./bench.sh
in the terminal , the programm gets compiled and execute for the parameters you set.
In this project , we are called to modify a producer-consumer solution that uses pthreads. The original program spawns one producer and one consumer thread , the original programm's code is locate in the prod-cons_default.c file . The purpose is the new program to be executed by p producer threads and q consumer threads. In the queue , the producer threads will add functions , using the struct workFunction in a FIFO queue and the role of the consumers threads is to execute these functions in parallel.
The struct workFunction is defined as :
struct workFunction {
void * (*work)(void *);
void * arg;
}
Furthermore we are asked to take measurements about the average waiting time of a workFunction item in the FIFO queue , and find for what number of consumers q , we get the smallest one .
Here are the tables with my measurements :
Measurements for QUEUESIZE =10 , time is in microseconds (usec) :
Measurements for QUEUESIZE =1000 , time is in microseconds (usec) :
The conclusion about the mean waiting time, that we get for the measurements in the tables above , is that:
1. For a constant number of producer threads p , as the number of consumers q is rising , the mean waiting number of a function is getting greater, until some point in reaches a small value (about 1.4-2 microseconds) and after by rising the number of consumer q we get some small differences.
2. Bigger QUEUESIZE means bigger queue and bigger mean waiting time ,as more functions can be in a pending state inside our FIFO queue
The conlusion number 1 , can be justified , also by the bellow graphs:
Graph for QUEUESIZE =10 :
Graph for QUEUESIZE =1000 :
The above conlcusions and measurements arise from the fact that we used functions in the myFunctions.c file that are not time costly , if we used "heavier functions", the average waiting time would be greater ,of course , but the conclusions about the relation between p,q and mean waiting time would stay about the same.