Title:An Optimal Iterative Placement Algorithm for PIR from Heterogeneous Storage-Constrained Databases

Abstract:We study private information retrieval (PIR) where a user privately downloads one of $K$ messages from $N$ databases (DBs) such that no DB can infer which message is being downloaded. Moreover, we consider the general case where DBs are storage constrained such that DB$_n$ can only store a $\mu[n]KL$ symbols where $0\leq \mu[n] \leq 1$ and $L$ is the number of symbols per message. Let $t= \sum_{n=1}^{N}\mu[n]$ be an integer, a recent work by Banawan et al. showed that the capacity of heterogeneous Storage Constrained PIR (SC-PIR) is $\left( 1+ \frac{1}{t} + \frac{1}{t^2} + \cdots + \frac{1}{t^{K-1}} \right)^{-1}$. However, an achievable, capacity achieving scheme was only developed for a network of $N=3$ DBs. In this paper, we propose an iterative placement algorithm for arbitrary $N$ which achieves heterogeneous SC-PIR capacity when $t$ is an integer. The algorithm defines storage contents of the DBs by assigning sets of sub-messages to $t$ DBs in each iteration. We show that the proposed placement algorithm converges within $N$ iterations and the storage placement requires at most $N$ sub-messages per message without considering the sub-message requirement for the delivery. Finally, we show that the proposed solution can be applied to the case of non-integer $t$ while still achieving capacity.