The exponential growth of real‑time sensor data, click‑stream logs, and IoT telemetry has intensified the need for storage systems that combine , low‑latency random access , and strong fault tolerance . In this paper we present FSDSS‑825 (Fault‑tolerant Scalable Distributed Streaming Storage, version 825) , a novel architecture that integrates log‑structured merge trees (LSM‑trees) with erasure‑coded sharding and deterministic replay to achieve linear scalability while guaranteeing exactly‑once semantics for streaming writes. We describe the design decisions, provide a formal model of consistency, and evaluate the system on a 128‑node cluster using the TPC‑DS and YCSB benchmarks. Results demonstrate up to 3.7× higher ingestion throughput and 2.4× lower tail latency compared with state‑of‑the‑art systems such as Apache Pulsar + BookKeeper and Apache Kafka + RocksDB, while sustaining a 99.999% durability guarantee under simultaneous node failures.
The remainder of the paper is organized as follows: Section 2 reviews related work, Section 3 presents the system architecture, Section 4 details the design of each component, Section 5 discusses the consistency model and failure handling, Section 6 reports experimental results, Section 7 provides a discussion of limitations, and Section 8 concludes with future directions. fsdss-825