EÆcient Pure-bu er Algorithms for Real-time Systems

We present several wait-free algorithms for implementing read/write bu ers in real-time systems. Such bu ers are commonly used in situations where newly-produced data values take precedence over older data, and hence older data can be overwritten. Each of our algorithms is a \pure-bu er" algorithm. In a purebu er algorithm, several bu ers are shared between the writer and reader processes, and a handshaking mechanism is employed that ensures that a writer never writes into a bu er that is concurrently being read by some reader. Each of our algorithms is optimized by taking characteristics of quantumand priority-based schedulers into account. When used to implement a B-word bu er that is shared across a constant number of processors, the time complexity for reading and writing in each of our algorithms is O(B), and the space complexity is (B). These complexity gures are obviously asymptotically optimal and are independent of the number of writer and reader processes. In contrast, all previously-published pure-bu er algorithms are limited to one writer process and have time and space complexity that is at least linear in the number of