Abstract: Decentralized tuple space implements tuple space model among a series of decentralized nodes and provides the global shared tuple repository. But, in spite of its advantages, large access latency is brought by the frequent remote operations and the conventional performance improving approach that dynamic moving tuples among nodes ceases to be effective if tuple’s locations are immutable, such as in LIME. In this study, a replication-based tuple distribution approach is proposed to improve performance of tuple access. A replicable decentralized tuple space model is presented, which gives semantics of tuple space primitives under replication. In this model, tuples are categorized as their use patterns and applied differentiated replication policies. According to the model, replication management modules are implemented based on LIME2. The experiment results showed that the read operation’s latency is obviously reduced. However, a large penalty is incurred for writing operations. Therefore, applying differentiated replication policies is absolutely necessary, because no single policy is suitable for all cases.