Atomic Snapshots from Small Registers

Abstract

Existing n-process implementations of atomic snapshots from registers use large registers. We consider the problem of implementing an m-component snapshot from small, Theta(log(n))-bit registers. A natural solution is to consider simulating the large registers. Doing so straightforwardly can significantly increase the step complexity. We introduce the notion of an interruptible read and show how it can reduce the step complexity of simulating the large registers in the snapshot of Afek et al. In particular, we show how to modify a recent large register simulation to support interruptible reads. Using this modified simulation, the step complexity of UPDATE and SCAN changes from Theta(n*m) to Theta(n*m+m*w), instead of Theta(n*m*w), if each component of the snapshot consists of Theta(w*log(n)) bits. We also show how to modify a limited-use snapshot to use small registers when the number of UPDATE operations is in n^{O(1)}. In this case, we change the step complexity of UPDATE from Theta((log(n))^3) to O(w + (log(n))^2*log(m)) and the step complexity of SCAN from Theta(log(n)) to O(m*w + log(n)).