SPDK
SPDK is a userspace storage stack commonly used as a kernel-bypass NVMe reference: it exposes polling, per-core queue pairs, and zero-copy-oriented paths while trading away parts of the conventional filesystem and kernel I/O interface.
是什么
In this corpus, SPDK is mainly a baseline or building block for high-performance local NVMe I/O. It makes the data path explicit and minimizes kernel crossings, so papers compare against it when they want to separate device throughput from page-cache, filesystem, or completion-path overhead.
It is not automatically the right deployment choice. Systems that retain POSIX filesystems, buffered I/O, or shared device access must either integrate only selected SPDK ideas or accept additional compatibility work.
关键观察 / 隐含假设
- 观察：userspace polling and per-core queues can approach device capability, but CPU ownership and workload contention determine whether polling remains desirable. DPAS-FAST26 contrasts polling, hybrid polling, and interrupts under contention.
- 观察：filesystem compatibility can be as important as raw I/O. uCache-FAST26 borrows SPDK-style NVMe ideas while retaining an ext4-assisted control path.
- 假设：a benchmark close to SPDK isolates software-stack overhead. uCache-FAST26 uses this comparison for its NVMe uStore, but its result is bounded by the specific random-read configurations.
演进时间线
- 2025 SOSP：Aeolia-SOSP25 — uses userspace I/O context in a storage-system design.
- 2026 FAST：uCache-FAST26 — combines SPDK-inspired data-path ideas with a unikernel cache and filesystem control path.
- 2026 FAST：DPAS-FAST26 — analyzes completion-mode trade-offs relevant to polling-based paths.
相关概念
相关论文
- uCache-FAST26 — compares uStore with SPDK while retaining filesystem compatibility.
- WSBuffer-FAST26 — studies buffered versus direct high-bandwidth SSD paths.
- DPAS-FAST26 — shows why completion mode depends on CPU contention and queue depth.
- RISTRETTO-FAST26 — places kernel-bypass paths in a cloud-local-storage evolution.