Optimal batch codes: Many items or low retrieval requirement

Abstract

Combinatorial batch codes were introduced by Ishai et al. [36th ACM STOC (2004), 262-271] and studied in detail by Paterson et al. [Adv. Math. Commun., 3 (2009), 13-27] for the purpose of distributed storage and retrieval of items of a database on a given number of servers in an economical way. A combinatorial batch code with parameters $n,k,m,t$ means that $n$ items are stored on $m$ servers such that any $k$ different items can be retrieved by reading out at most $t$ items from each server. If $t=1$, this can equivalently be represented with a family $\mathcal F$ of $n$ not necessarily distinct sets over an $m$-element underlying set, such that the union of any $i$ members of $\mathcal F$ has cardinality at least $i$, for every $1\le i\le k$. The goal is to determine the minimum $N(n,k,m)$ of $\sum$$F\in\mathcal F$$|F|$ over all combinatorial batch codes $\mathcal F$ with given parameters $n,k,m$ and $t=1$. We prove $N(n,k,m)= (k-1)n- \lfloor \frac{(k-1){m \choose k-1}-n}{m-k+1} \rfloor$ for all ${m\choose k-2} \le n \le (k-1){m\choose k-1}$.Together with the results of Paterson et al. for $n$ larger, this completes the determination of $N(n,3,m)$.We also compute $N(n,4,m)$ in the entire range $n\ge m\ge 4$.Several types of code transformations keeping optimality are described, too.