摘要：We present two wait-free algorithms for simulating an l-bit single-writer register from k-bit single-writer registers, for any k >= 1. Our first algorithm has big-theta(l/k) step complexity for both Read and Write and uses big-theta (4^(l-k)) registers. An interesting feature of the algorithm is that Read operations do not write to shared variables. Our second algorithm has big-theta (l/k + (log n)/k) step complexity for both Read and Write, where n is the number of readers, but uses only big-theta (nl/k + n(log n)/k) registers. Combining both algorithms gives an implementation with big-theta (l/k) step complexity using big-theta (nl/k) space for any 1 <= k < l. We also prove that any implementation with big-O (l/k) step complexity for Read requires big-omega (l/k) step complexity for Write. Since reading l-bits requires at least ceiling(l/k) reads of k-bit registers, our lower bound shows that our implementation is step optimal.
关键词：atomic register; regular register; wait-free implementation; single writer; optimal
