ESBT: new CRDT cuts memory 75% and speeds edits 88%

Conflict-free Replicated Data Types (CRDTs) are the backbone of modern decentralized collaborative editing tools, letting replicas apply concurrent operations in any order and still converge. But classic sequence CRDTs like Logoot and LSEQ suffer from uncontrolled identifier growth: every insertion adds a unique, increasingly large tag, so memory consumption balloons during long, high-concurrency sessions. This forces editors to either reset state or degrade performance.

ESBT (Extended Stern-Brocot Tree) attacks this with a mathematically grounded identifier allocator based on the Stern-Brocot tree, producing dense, deterministic, and compact identifiers that stay small regardless of edit history. The paper reports that under workloads of up to 100,000 operations across 50 simulated sites, ESBT improved responsiveness by 28-88% for pure insertions and 59-74% for mixed insertion/deletion workloads, while cutting identifier memory use by 50-75% compared to Logoot and LSEQ. In a stress-test of adversarial middle-insertions (10,000 ops), ESBT was 86.53% more responsive and reduced identifier size by 92.81%. That means real-time editors can scale to far larger groups and longer sessions without the usual memory blowup or latency creep.