| 王雪* **,李文青* **,张福新* **,王剑* **,敖琪***.基于紧耦合内存复制加速器的Java垃圾收集优化方法[J].高技术通讯(中文),2025,35(7):687~697 |
| 基于紧耦合内存复制加速器的Java垃圾收集优化方法 |
| An optimization method for Java garbage collection based on tightly coupled memory copy accelerator |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2025. 07. 002 |
| 中文关键词: 垃圾收集;对象移动;紧耦合内存复制加速器;细粒度并行;Java 虚拟机 |
| 英文关键词: garbage collection,object movement,tightly coupled memory copy accelerator,fine-grained parallelism,Java virtual machine |
| 基金项目: |
| 作者 | 单位 | | 王雪* ** | (* 处理器芯片全国重点实验室(中国科学院计算技术研究所) 北京 100190)
(** 中国科学院大学 北京 100049)
(*** 龙芯中科技术有限公司 北京 100190) | | 李文青* ** | | | 张福新* ** | | | 王剑* ** | | | 敖琪*** | |
|
| 摘要点击次数: 349 |
| 全文下载次数: 545 |
| 中文摘要: |
| 垃圾收集(garbage collection,GC)是 Java 虚拟机(Java virtual machine,JVM)的重要组成部分,它在确保内存安全性的同时,提高了应用的开发效率。 然而,垃圾收集会导致应用程序吞吐率下降或长时间暂停。 对象移动开销大是影响主流垃圾收集器效率的常见问题,本文提出了一种 Java 垃圾收集优化方法。 此方法采用紧耦合内存复制加速器来加速垃圾收集过程中耗时的对象移动,并将对象复制和遍历任务细粒度并行执行,从而充分发挥将对象复制任务卸载到加速器执行的性能优势。 实验结果表明,此方法能够有效提高垃圾收集效率,减少应用程序暂停时间。 在 DaCapo 和 SPECjvm 2008 基准测试集中,与默认的 G1(garbage-first)垃圾收集器相比,此方法垃圾收集吞吐量分别平均提高8. 59% 和8. 81% , 最高可达 74. 69% 。 此外, 其暂停时间平均分别减少了 5. 08% 和5. 96% ,最多减少了 40. 44% 。 |
| 英文摘要: |
| Garbage collection (GC) is a crucial component of the Java virtual machine(JVM). It ensures memory safety while enhancing the efficiency of application development. However,garbage collection often leads to reduced application throughput or prolonged pauses. A common challenge affecting the effectiveness of mainstream garbage collectors is the significant overhead associated with object movement. This paper proposes an optimization method for Java garbage collection. The method employs a tightly coupled memory copy accelerator to expedite the timeconsuming object movement during garbage collection. It achieves fine-grained parallel execution of object copy and traversal tasks,fully capitalizing on the performance benefits of offloading object copy tasks to the accelerator. Experimental results demonstrate that this method effectively improves garbage collection efficiency and reduces application pause times. Compared with the default G1 GC,the garbage collection throughput of this method for the DaCapo and SPECjvm2008 benchmark suites is increased by an average of 8. 59% and 8. 81% ,respectively,with a maximum improvement of up to 74. 69% . Additionally, its pause time is decreased by an average of 5. 08% and 5. 96% respectively,with a maximum reduction of 40. 44% . |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
