چكيده به لاتين
Tail latency is problematic for several reasons. Interactive services can struggle to provide complex functionality under the strict latency budgets required to ensure responsiveness. Under high degrees of parallelism, poor tail latency will impact most user requests because user’s request does not complete until the slowest of these sub-requests has finished. We show how hardware, operating system, and application-level design and configuration choices introduce latency variability. In this thesis, we focus on operating system, especially scheduler. Scheduler plays an important role in I/O intensive applications. This scheduler is well designed for managing resources. The default Completely Fair Scheduler guarantees fairness for resources. Scheduler has a duty to set up justice for tasks to access processor. Given the difference in tasks, CPU time should be shared across all tasks. In this case, starvation will not happen. Our proposed solution in this thesis is to change the behavior of the operating system scheduler, which has a fair behavior against the I/O intensive applications. In this solution, unlike other existing solutions, we are looking for changes to the existing scheduler, which, along with the benefits of this scheduler, also resolves the problem of the tail latency. In fact, by changing Linux's default scheduling behavior in I/O intensive applications, by modifying the definition of fairness; we changed the default Linux scheduler.
Keywords: Scheduler, Linux, I/O Process, Completely Fair Scheduler, CFS
