lateFs

lateFs is a proposed distributed file system designed to reduce latency in cloud-scale storage by employing late binding of data placement and metadata. The central idea is that the system buffers writes and defers the final association of data blocks with a file namespace until the blocks are actually accessed, enabling low-latency acknowledgement of writes in bursty workloads.

Architecture and data model: lateFs separates metadata from data and uses a scalable metadata service to track tentative block mappings. Data blocks are stored in an object store, and a write-ahead log ensures durability. The binding map records provisional placements and is reconciled in the background as activity occurs.

Operation: On write, lateFs appends to the log and returns success; actual data placement is decided later. On read, the system consults a local cache or the tentative mapping; if a block is not yet bound, it fetches from storage and materializes it, updating the map. Background processes perform cleanup, replication, and compaction while preserving eventual consistency.

Features and use cases: lateFs supports versioning and simple snapshots, sparse files, and strong durability through replication. It is aimed at workloads with bursty write traffic, large-scale content delivery, and cloud-native applications that can tolerate occasional read-after-write latency in exchange for low write latency.

Status: lateFs exists primarily as a conceptual design and in early-stage prototypes described in research discussions. It is not part of any standard or major production distribution, and practical deployments would require careful consideration of consistency, recovery, and POSIX compatibility.