Chapter 7:

1. Any data structure designed for thread safety must have a transactional interface: every operation must either not change the state of the data structure or transform it from one well-defined state to another well-defined state.
2. This comes to the general observation of the performance of concurrent code: the more shared variables there are, the slower the code is. A complex data structure usually needs more data shared between threads that access it concurrently. In addition, there are simple algorithms (some are wait-free) that allow limited thread-safe operations on the data structures.
3. With an efficient lock, a lock-guarded data structure is not necessarily slower. Often, it is faster. Again, it comes to how many variables are shared: a lock-free scheme that requires multiple atomic variables may be slower than a single lock. We also have to consider the locality of the access: if the data structure is accessed in one or two places (like a queue), the lock...