cloud storage architects

Abstract: Recently, many blockchain-based medical data sharing systems using off-chain storage have been proposed. However, these systems have some drawbacks such as being incompletely decentralized, low adaptability to cloud storages, high cost, and low security. It is non-trivial to address the drawbacks due to the transparency, immutability, and decentralization of a blockchain platform. Our goal is to design a decentralized medical data sharing system with off-chain storage that addresses all the above drawbacks while satisfying data protection and privacy laws. We propose the VASCO (View-based AccesS COntrol) system that does not rely on any central server including a key generation one, not rely on any special cloud storage including IPFS, not use a secure channel, encrypts both data and metadata, and supports end-to-end encryption. We achieve it by designing eight chaincodes rigorously based on view-based access control. For a medical data ecosystem that consists of many stakeholders who employ different data standards, storage formats, and transmission protocols, we present how typical flows of medical data in the ecosystem can be achieved by using the proposed chaincodes with interoperability. We showed that VASCO is superior to the existing systems in terms of both functionality and regulatory compliance. We also tested the scalability of VASCO using a blockchain of 21 peer nodes with different specifications and verified interoperability through 30 combinations of standards, formats, and protocols. VASCO is a promising medical data sharing system that can provide secure and efficient sharing of medical data while preserving patient privacy and satisfying regulatory compliance with interoperability. The result of this study provides a potential to improve patient outcomes and advance medical research for the medical data ecosystem.

Abstract: Long-context LLM serving for agentic, coding, and document workloads reuses prefixes whose key-value (KV) cache outgrows GPU and host memory, putting storage I/O on the critical path of the inference step for retrieving prefixes. A single node-local NVMe absorbs all retrievals, while the parallel file system or cloud file store that nodes already mount carries almost no inference traffic. We cast prefix retrieval as aggregate-bandwidth scheduling across these tiers and present KVStream, a KV cache engine that batches a step's reads, defers writes behind the forward pass, replicates each chunk on every tier, and lets each tier adaptively claim the next unread chunks once it is free. Our performance model shows this routing attains the aggregate-bandwidth bound without prediction or tuning, whereas any static split misses it by a margin that grows as per-tier bandwidths drift or fluctuate. It also predicts that the time-to-first-token (TTFT) gain grows with load, as faster retrieval drains the queue, and disappears when host memory or arrival order keeps reuse in cache. Plugged into LMCache and evaluated on four models, six datasets, and up to 128 GPUs, KVStream retrieves within 5% of the bound, delivers up to 1.49$\times$ higher offline throughput than LMCache, and lowers mean TTFT by up to 6.1$\times$ under bursty arrivals.

Abstract: Most databases, including PostgreSQL, RocksDB, and recent AI KV-cache middleware, rely on buffered I/O, delegating write-back to the Linux kernel. On the distributed block storage standard in the cloud, this delegation inherits a hidden bottleneck: each device is drained by a single kernel flusher thread over a high-latency, shallow-queue path. When the drain falls behind, dirty throttling pauses write() system calls, and even reads that must evict dirty pages stall. These stalls are invisible to iostat and every standard counter. This poster observes the stall from inside the kernel, using the multi-volume data placement proposed in SteelDB as the experimental lever. eBPF probes on writeback and block tracepoints separate write-back by issuing context and count every throttle pause. Across three configurations with identical provisioned IOPS and bandwidth but 1, 2, and 4 devices, we show that adding drains, not bandwidth, cuts throttle pauses by 70%, reduces maximum transaction latency by 59%, and raises throughput by 23%.