Impact of priority bonuses of Inter-Core Aggregation Scheduler on a commodity CMP platform

This paper investigates the effect of changing priority bonuses of Inter-Core Aggregation Scheduler (IAS). IAS is a kernellevel thread scheduler for commodity Chip Multi-Processing (CMP) platforms. To utilize a shared cache and memory on a CMP platform to enhance the performance, IAS executes sibling threads, kernel-level threads sharing the same memory address space, simultaneously on different Cores. To enable this scheduling mechanism, IAS uses the priority bonuses which dynamically enhance the priority of sibling threads in the runqueues of all Cores in scheduling next threads inside the kernel. We evaluate the impact of changing priority bonuses in executing RUBiS benchmark, a Web application benchmark modeling eBay.com. According to our experimental result, IAS enhances the performance of RUBiS benchmark with every priority bonus we applied. We also show that the effect of IAS is influenced by the priority bonuses and we can enhance the effect of IAS by tuning the priority bonuses properly.