Commensal cuckoo

Abstract

We present commensal cuckoo,* a secure group partitioning scheme for large-scale systems that maintains the correctness of many small groups, despite a Byzantine adversary that controls a constant (global) fraction of all nodes. In particular, the adversary is allowed to repeatedly rejoin faulty nodes to the system in an arbitrary adaptive manner, e.g., to collocate them in the same group. Commensal cuckoo addresses serious practical limitations of the state-ofthe- art scheme, the cuckoo rule of Awerbuch and Scheideler, tolerating 32x--41x more faulty nodes with groups as small as 64 nodes (as compared to the hundreds required by the cuckoo rule). Secure group partitioning is a key component of highly-scalable, reliable systems such as Byzantine faulttolerant distributed hash tables (DHTs).