This is the kernel source that can be used for GiantVM environments with optimized performance. It includes the following improvements:
- Realignment of kernel data to avoid false sharing
- Implementation of a per-node RW semaphore for mmap_sem, which reduces DSM overheads caused by the contention for the semaphore variable
- To use this feature, you should enable MMSEM and configure MMSEM_NODE(default:4) and MMSEM_CPU(default:16) according to your system
Contributors: MemOS project members in Electronics and Telecommunications Research Institute (ETRI), South Korea
- Baik Song An (bsahn@etri.re.kr)
- Myung Hoon Cha (mhcha@etri.re.kr)
- Hong Yeon Kim (kimhy@etri.re.kr)
- Sang-Min Lee (sangmin2@etri.re.kr)
License
- GPL v2.0