主页
加入实验室
团队成员
教师信息
学术活动
内存管理
内存墙一直是各类计算系统中的瓶颈,从移动计算系统、PC、服务器,到智能系统计算系统(大模型、GPU)。
现有工作
优化对象:
面向NVM、hybrid memory、大数据的GC
优化目标:
相关文献
Bridging the performance gap for copy-based garbage collectors atop non-volatile memory. (EuroSys ‘21). 343–358. https://doi.org/10.1145/3447786.3456246 (面向NVM的GC)
Reliability-aware Garbage Collection for Hybrid HBM-DRAM Memories. ACM Trans. Archit. Code Optim. 18, 1, Article 10 (March 2021), 25 pages. https://doi.org/10.1145/3431803 (面向hybrid memory的GC)
Transparent and lightweight object placement for managed workloads atop hybrid memories. (VEE 2022). 72–80. https://doi.org/10.1145/3516807.3516822 (面向hybrid memory的GC)
Low-latency, high-throughput garbage collection. (PLDI 2022). 76–91. https://doi.org/10.1145/3519939.3523440 (tracing + RC混合GC,综合高吞吐和低延迟)(有源码)
Deconstructing the garbage-first collector. (VEE ‘20). 15–29. https://doi.org/10.1145/3381052.3381320 (G1 GC的评估分析)
Better Understanding the Costs and Benefits of Automatic Memory Management. (MPLR ‘22). 29–44. https://doi.org/10.1145/3546918.3546926 (综述)
BestGC: An Automatic GC Selector Software. (MPLR ‘22). 144–146. https://doi.org/10.1145/3546918.3560804(综述)
Analysis of Garbage Collection Patterns to Extend Microbenchmarks for Big Data Workloads (面向大数据的GC)
Unified Holistic Memory Management Supporting Multiple Big Data Processing Frameworks over Hybrid Memories. ACM Trans. Comput. Syst. 39, 1–4, Article 2 (November 2021), 38 pages. https://doi.org/10.1145/3511211 (面向大数据的GC)
TeraHeap: Reducing Memory Pressure in Managed Big Data Frameworks. (ASPLOS 2023).694–709. https://doi.org/10.1145/3582016.3582045(面向大数据的GC)
Jade: A High-throughput Concurrent Copying Garbage Collector. (EuroSys ‘24). 1160–1174. https://doi.org/10.1145/3627703.3650087 (高吞吐GC)
Let It Go: Relieving Garbage Collection Pain for Latency Critical Applications in Golang.(HPDC ‘23).169–180. https://doi.org/10.1145/3588195.3592998(低延迟GC)
User-Centric Approaches to Garbage Collector Selection and Heap Size Optimization for Java Applications (博士论文)
可能尝试的方向
针对应用特点(比如移动应用、云服务),提出高效的管理方法
针对应用的目标要求(比如吞吐量、延迟),提出高效的管理方法
高性能的RC GC方法