Guangyan Zhang

Enterprise SSDs integrate substantial computing resources (e.g., ARM processor and onboard DRAM) to handle I/O bursts. However, these resources significantly raise SSD monetary cost and suffer severely underutilized in JBOF deployments due to the sporadic nature of I/O bursts. In this paper, we p...

Our production experience in large-scale data-processing pipelines shows that, despite its attractive programming model, memory-mapped I/O (_mmap-IO_) on distributed file systems (DFS) exhibits severe operational pathologies—livelocks during write-heavy phases, chronically low multi-threaded thro...

Wide-stripe erasure codes, with tens to over a hundred data chunks per stripe, offer high reliability at low storage overhead. Existing wide-stripe designs that are based on scalar codes (e.g., LRCs in Google and Azure) reduce _repair traffic_ but increase _storage overhead_. Although vector code...

In interactive LLM serving, historical key–value tensors (KVs) of multi-round conversations are often cached in a two-tier storage system consisting of host memory and SSDs, which provides large capacity at low cost. However, loading KVs from two-tier storage in existing approaches increases serv...

Minimum Storage Regenerating (MSR) codes have strong potential for building efficient and reliable storage systems due to their excellent fault tolerance and low repair bandwidth. However, to meet MSR code constraints and optimize storage performance, systems often adopt large chunk sizes. This l...