I.K.T. Tan, C.S. Wong, J.W. Lam, and R.D. Kumari (Malaysia)
Operating system scheduler, parallel systems, fairness measurement and benchmarking.
With more software architects and developers coding for parallel execution, how fair tasks are scheduled by the operating system becomes an important criteria. Software code may comprise of small sections that are parallelizable and every possible performance gain should be exploited by the software developer. In order to exploit fine grain parallelism, software developers need the confidence that the operating system is able to fairly schedule their parallelized tasks. Most schedulers attempt to allocate resources to tasks fairly based on the task’s priority. However, this fairness cannot be achieved in an ideal manner and hence it is only an approximate fairness. Actual experience with various schedulers varies and currently, there is no tool to qualitatively measure and compare them. This paper presents a tool to measure fairness and provides an intuitive representation of the results through the comparison of two different kernel schedulers of the open source Linux operating system.
Important Links:
Go Back
