A REPORT ON Analysis of Reference Counting Algorithms for RCU based Lock-free Data Structures in Linux Kernel

IBM for giving me the opportunity of doing my Practice School-II project at IBM. I would also like to thank Prof. I am indebted my mentor, Mr. Srivatsa Vaddagiri for his constant help, technical guidance and inspiration throughout the work. I express my sincere gratitude to Mr. their help during the course of work. Finally, I would also like to thank the entire LTC team for their support and guidance. I am also grateful to Ms. Rita Gopalan, PS-II faculty for her constant guidance and invaluable suggestions. Abstract : With the advent of shared-memory parallel (SMP) systems, an additional level of complexity was added to locking primitives. Read-copy update(RCU) [1, 2] like many other synchronization primitives was hence developed as a reader-writer synchronization mechanism where readers can incur zero synchronization overhead. RCU [4] provides improved performance in the case of read-mostly data structures. RCU is implemented in the linux kernel(eg. dcache, route cache) in different forms. However, most of them use reference counting [6] for keeping track of the references to (iv) the shared data structure. Some of these RCU algorithms and the different atomic synchronization operations that they use are analyzed during the course of this work. The lookup performance of these algorithms is measured under different scenarios and is compared with respect to each other.