Storage Allocation for Multi-Class Distributed Data Storage Systems

TL;DR

This paper investigates optimal storage allocation strategies in multi-class distributed storage systems to maximize data recovery success probabilities while satisfying quality of service requirements, proposing minimal spreading allocation as an effective solution.

Contribution

It introduces the minimal spreading allocation approach and demonstrates its near-optimal performance through theoretical bounds and simulations in complex storage scenarios.

Findings

Minimal spreading allocation approaches optimal performance in many cases.

Optimal storage allocation can be approximated using upper bounds.

Simulations confirm the effectiveness of the proposed method.

Abstract

Distributed storage systems (DSSs) provide a scalable solution for reliably storing massive amounts of data coming from various sources. Heterogeneity of these data sources often means different data classes (types) exist in a DSS, each needing a different level of quality of service (QoS). As a result, efficient data storage and retrieval processes that satisfy various QoS requirements are needed. This paper studies storage allocation, meaning how data of different classes must be spread over the set of storage nodes of a DSS. More specifically, assuming a probabilistic access to the storage nodes, we aim at maximizing the weighted sum of the probability of successful data recovery of data classes, when for each class a minimum QoS (probability of successful recovery) is guaranteed. Solving this optimization problem for a general setup is intractable. Thus, we find the optimal storage allocation when the data of each class is spread minimally over the storage nodes, i.e. minimal spreading allocation (MSA). Using upper bounds on the performance of the optimal storage allocation, we show that the optimal MSA allocation approaches the optimal performance in many practical cases. Computer simulations are also presented to better illustrate the results.