Fast Collect in the absence of contention

We present a generic module, called Fast Collect. Fast Collect is an implementation of Single-Writer Multi-Reader (SWMR) Shared-Memory in an asynchronous system in which a processor updates its cell and then reads in any order all the other cells. Our simple implementation of Fast Collect uses some MultiWriter, Multi-Reader (MWMR) variables and one local boolean per processor, such that eventually, in the absence of contention, i.e. if only a single processor repeatedly performs collect, the amortized cost per each collect is a constant. At the example of Disk Paxos we show, how Fast Collect can be used as a building block in wait-free algorithms.