ABS: An Availability-Balanced Data Placement Strategy to Improve Overall Data Availability in Decentralized Storage

Decentralized storage systems rely on heterogeneous edge nodes and therefore often suffer from weak availability guarantees. Although redundancy mechanisms such as replication and erasure coding can improve data availability, the overall data availability still depends heavily on how replicas or coded blocks are placed across nodes with different availability levels. Existing placement approaches can be broadly divided into two categories: random strategies and greedy strategies. Random strategies ignore node heterogeneity, while greedy strategies tend to over-rely on a limited number of highly available nodes and decline as data loading proceeds. In this paper, we develop an availability analysis model for decentralized storage under both replication and erasure coding. We find that, in most cases, balancing availability across data items helps improve the overall data availability. Based on these insights, we propose ABS, an A vailability- B alanced placement S trategy. ABS uses grouped bidirectional matching to balance the data-item availabilities. More specifically, under replication, it steers each data item toward the standard value; under erasure coding, it adopts a variance-greedy final selection rule. We further incorporate several complementary techniques to optimize read efficiency, load balance, and migration cost. Prototyping and simulation-based experiments show that ABS consistently outperforms random and greedy baselines in average data availability under both replication and erasure coding settings.