System and method that facilitate the prediction performance of the multithreaded application in presence of resource bottlenecks

摘要

A system for determining the performance of a software application multithreading in the presence of one or more resource bottlenecks, system comprising: a processor (202); and a memory (208) coupled to the processor (202), wherein the processor (202) is capable of executing a plurality of modules stored in memory (208), and wherein the plurality of modules comprises: a module (212) representation configured to represent a network of queues hardware and network queues software resources used to implement the application software multithreading, where networks queues represents contention for resources, and wherein the resources include hardware resources and software resources in networks where queues are used to check the number of threads that access the application software multithreading when there is a bottleneck of resources; a module (214) computer configured to run an iterative technique to a predetermined value of the service request to develop a relationship between the network queues software and network queues hardware, wherein the ratio comprises using software resources regarding hardware resources based on the predetermined value of the service request, and wherein the execution of the iterative technique comprises: obtaining a number of blocked threads in a critical section of the resources software based on the predetermined value of the service request, wherein the threads blocked in the critical section are threads waiting to enter the critical section; obtain a residence time in each of the hardware resources based on the predetermined value of the service request; update a service request critical section of a current interaction for a service request critical section updated to reflect a wait for hardware resources using the residence time, where the service request critical section is updated iteratively to take account of resource contention on hardware resources based on time of residence; comparing the number of blocked the current iteration number blocked a previous iteration in order to check whether the number of blocked threads of the current interaction is higher than the number of blocked from the previous iteration threads threads threads, and check if the service request updated critical section is higher than the critical service request section of the previous iteration; and identifying a difference between the number of blocked threads in a critical section for the current iteration and the number of blocked in the critical section in the previous iteration threads; and repeatedly executing the iterative technique 1) if the difference in the number of blocked threads is greater than a predefined limit, 2) if the number of blocked threads the current iteration is higher than the number of threads blocked prior iteration and 3) if the service request updated critical section is greater than the critical service request section of the previous iteration; otherwise obtain performance metrics with the service request critical section updates for application software multithreading when the difference in the number of blocked threads is less than the predefined limit, wherein the predefined limit indicates that the difference in the number of blocked threads it has converged on a number of iterations of the iterative technique, thereby determining a yield of the software application multithreading.