To overcome the limitations (e.g., vulnerability to central server failure) of current existing microblogging systems (e.g., Twitter, and Weibo), researchers have proposed several decentralized microblogging solutions. However, these solutions also have several issues: (i) they were not designed for (and hence cannot handle) scenarios where massive correlated failures occur; or (ii) their delivery rates were not significantly high (i.e., lower than 85%); or (iii) their working mechanisms were not evaluated on partitioned networks based on a ground-truth dataset with a real publisher–subscriber distribution. In this paper, we propose SAND: a microblogging system builds upon an overlay where users have private IP connections to their social friends, enabling trusted social communication even on partitioned networks. We assess SAND through a simulation-based analysis considering a real-world dataset crawled from Twitter, as well as a synthetic dataset that provides high flexibility to tune network parameters. The results show that SAND is feasible and efficient, also in the case of network partitions. For example, in a partitioned network where 98% of network nodes fail, by using SAND-SN, peers are able to effectively follow each other’s updates (i.e., 100% delivery rate). On average, the overhead is 1.9 copies, 22 network sends and 6.6 network receives per message.