Multi-threaded applications are used to share data structures and to access the memory in a concurrent manner. This access is typically protected by mutual exclusion paradigms, such as mutexes and semaphores, that may cause sequential bottlenecks in a time-critical software. Thanks to special processor instructions, it is possible to avoid classical locking and perform lock-free operations on shared data structures. This paradigm is extensively used in modern kernels and in some programming languages (e.g. the C++11 standard library).
The talk introduces this topic by discussing aspects of the UPMARC summer school course, in particular by providing some use-case examples and by discussing the advantages and disadvantages of the approach.