GQoSMT: On Guaranteeing the Quality of Service Requirements of Simultaneous Multithreading Processors

Guaranteeing the quality of service of a running thread on a Simultaneous Multithreading processor is one of the most challenging issues since these processors allow sharing of many datapath resources, including the Issue Queue, the Reorder Buffer, the Load/Store Queue, the branch prediction circuit, and multiple levels of the caches among multiple threads. In this study, we apply machine learning techniques to accurately predict the instant quality of service of a target thread so that we can transfer just a sufficient amount of shared resources to keep its quality of service stable at an expected level. Our test results show that our proposed prediction model gives only around 3% deviation from the target quality of service level, on average, whereas an earlier prediction model gives more than a 13% deviation.