Cloud object storage systems achieve scalability and fault isolation by sharding metadata across independently managed domains. While most metadata operations naturally scale under this model, rename remains a fundamental challenge, as it requires atomicity and strong consistency across multiple metadata owners. This challenge is exacerbated when source and destination metadata reside in different shards, operate without shared storage, and cannot depend on global locks or global clocks.
This talk presents the design of a distributed rename protocol for cloud-scale object storage systems, enabling atomic renames across independent metadata domains. The protocol models rename as a recoverable, multi-stage state machine rather than a monolithic transaction. It relies on explicit transaction tracking, shard-local invariants, and idempotent transitions, allowing failures and concurrent operations to safely drive in-progress renames toward completion or abort without centralized coordination.
Speaker(s): Amber
Agenda:
6:25 to 6:35 PM – Open for participants to enter and network.
6:35 to 6:40 PM – IEEE LM and CTCN Business meeting and to introduce speaker.
6:40 to 7:55PM – Formal Program and Q&A.
Virtual: https://events.vtools.ieee.org/m/550647