一句话：提出 EXIST，用现成硬件追踪实现数据中心服务内追踪，将开销降至千分位并保持高准确度。
现代云数据中心在线应用日益复杂，性能异常更难定位。精确推理事件因果既需 RPC 级跨服务追踪，也需服务内执行追踪，但现有服务内追踪平均开销约 5–10%，最坏达 18%，难以在共享且高负载数据中心实用。本文提出 EXIST，一个基于现成硬件追踪能力的高效服务内追踪系统。它通过三个协同模块降低开销：消除昂贵的追踪控制操作；依据应用状态动态分配有限追踪缓冲区；利用集群级编排优化追踪覆盖。作者在基准和真实应用上评估，EXIST 较现有技术效率提升 2–10 倍，与穷尽追踪参考相比准确率超过 90%，支持更可解释的数据中心管理。
The complexity of online applications is rapidly increasing, bringing more sophisticated performance anomalies in today's cloud datacenter. To fully understand application behaviors, we should obtain both inter-service communication data via RPC-level tracing and intra-service execution traces via application-level tracing to precisely reason about event causality. However, the average time overhead of existing intra-service tracing schemes on the traced applications is generally about 5-10%, possibly reaching 18% in the worst case. To realize practical intra-service tracing in shared and stressed datacenters, one must achieve extreme tracing efficiency with an overhead at the per-mille level. In this work, we present EXIST, an extremely efficient intra-service tracing system based on off-the-shelf hardware tracing capabilities. EXIST consists of three cooperative modules to pursue optimal trade-offs towards extremely low overhead. Firstly, it identifies and eliminates costly tracing control operations to guarantee the performance of the observed applications. Secondly, it allocates limited trace buffer space dynamically based on application status. Thirdly, it optimizes the trace coverage with cluster-level orchestration. We implement and evaluate EXIST on benchmark and real-world applications thoroughly. EXIST achieves 2-10× efficiency improvements compared to existing techniques and over 90% accuracy compared to exhaustive tracing reference. With extremely efficient intra-service tracing observability, we can achieve more explainable datacenter management.
@inproceedings{0003HLLLXYZWYCG25,
title = {EXIST: Enabling Extremely Efficient Intra-Service Tracing Observability in Datacenters},
author = {Xinkai Wang and Xiaofeng Hou and Chao Li and Yuancheng Li and Du Liu and Guoyao Xu and Guodong Yang and Liping Zhang and Yuemin Wu and Xiaopeng Yuan and Quan Chen and Minyi Guo},
booktitle = {ASPLOS},
year = {2025},
doi = {10.1145/3676641.3716283}
}