--- title: ATC 2024 Digest venue: ATC year: 2024 date: '2024-01-01' tags: [] paper_count: 12 draft: false --- 12 papers selected. --- ### FetchBPF: Customizable Prefetching Policies in Linux with eBPF *Xuechun Cao, Shaurya Patel, Soo-Yee Lim, Xueyuan Han *et al.** **TL;DR** — Extends eBPF into the page-fault / prefetch path, giving user-space programs a safe, low-overhead hook to install custom hardware-prefetch policies without kernel modifications. --- ### Fast (Trapless) Kernel Probes Everywhere *Jinghao Jia, Michael V. Le, Salman Ahmed 0001, Dan Williams 0001 *et al.** **TL;DR** — Eliminates the trap-based overhead of kprobes by using binary rewriting to instrument kernel functions at near-zero cost, enabling always-on production tracing. --- ### FBMM: Making Memory Management Extensible With Filesystems *Bijan Tabatabai, James Christopher Sorenson III, Michael M. Swift* **TL;DR** — Proposes delegating Linux virtual-memory management to filesystem drivers via a clean abstraction layer, opening the OS memory subsystem to the same extensibility that filesystems enjoy. --- ### Limitations and Opportunities of Modern Hardware Isolation Mechanisms *Xiangdong Chen, Zhaofeng Li 0004, Tirth Jain, Vikram Narayanan *et al.** **TL;DR** — Empirically evaluates Intel MPK, RISC-V sPMP, and related primitives and shows that their performance and safety properties diverge sharply from vendor claims, motivating a fresh look at hardware-assisted compartmentalization. --- ### FastCommit: resource-efficient, performant and cost-effective file system journaling *Harshad Shirwadkar, Saurabh Kadekodi, Theodore Y. Ts'o* **TL;DR** — Redesigns ext4 journal commits to write only changed metadata deltas rather than full blocks, slashing journaling overhead while preserving crash consistency — and is already deployed in the Linux kernel. --- ### ZMS: Zone Abstraction for Mobile Flash Storage *Joo Young Hwang, Seokhwan Kim, Daejun Park 0002, Yong-Gil Song *et al.** **TL;DR** — Introduces a zone-based abstraction that exposes the append-only write semantics of mobile UFS flash to the OS, cutting write amplification and GC overhead by an order of magnitude on mobile workloads. --- ### Ethane: An Asymmetric File System for Disaggregated Persistent Memory *Miao Cai 0001, Junru Shen, Baoliu Ye* **TL;DR** — Proposes a split-path file system design where reads bypass the server entirely and writes use lightweight logging, achieving near-DRAM read latency on disaggregated persistent memory. --- ### StreamCache: Revisiting Page Cache for File Scanning on Fast Storage Devices *Zhiyue Li, Guangyan Zhang* **TL;DR** — Demonstrates that the Linux page cache becomes a bottleneck — not a benefit — for sequential scans on NVMe SSDs, and replaces it with a lightweight streaming buffer that halves latency for analytical workloads. --- ### PeRF: Preemption-enabled RDMA Framework *Sugi Lee, Mingyu Choi, Ikjun Yeom, Younghoon Kim* **TL;DR** — Adds fine-grained preemption to RDMA by intercepting work-queue operations at the NIC driver level, enabling priority isolation for latency-sensitive RPC traffic sharing a fabric with bulk transfers. --- ### OSMOSIS: Enabling Multi-Tenancy in Datacenter SmartNICs *Mikhail Khalilov, Marcin Chrapek, Siyuan Shen, Alessandro Vezzu *et al.** **TL;DR** — Designs an OS-like resource manager for SmartNIC compute and memory that enforces tenant isolation and QoS, showing that shared-SmartNIC deployments are practical without sacrificing performance. --- ### mmTLS: Scaling the Performance of Encrypted Network Traffic Inspection *Junghan Yoon, Seunghyun Do, Duckwoo Kim, Taejoong Chung *et al.** **TL;DR** — Splits TLS session state across multiple cores using a novel sharding scheme, removing the per-connection serialization bottleneck in middlebox TLS inspection and achieving near-linear multi-core scaling. --- ### UniMem: Redesigning Disaggregated Memory within A Unified Local-Remote Memory Hierarchy *Yijie Zhong, Minqiang Zhou, Zhirong Shen, Jiwu Shu* **TL;DR** — Collapses the local/remote memory distinction into a single unified hierarchy with a new page-placement runtime, achieving transparent memory disaggregation with significantly lower tail latency than prior CXL-based approaches.