CAREER: Advanced Methods for Post-Deployment Debugging

职业：部署后调试的高级方法

• 批准号: 0953478
• 负责人: Benjamin Liblit
• 金额: $ 50万
• 依托单位: University of Wisconsin-Madison
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2010
• 资助国家: 美国
• 起止时间: 2010-03-15 至 2016-02-29
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0953478&HistoricalAwards=false
• 关键词: CAREER; Advanced; Methods; Post; Deployment

Computer technology is rapidly permeating all spheres of society. A computer system that affects the lives of thousands or millions of people creates a massive community of users who have an interest in the correct behavior of that system. Widespread interconnectivity means that we now have the ability to tap this potential.This work confronts the challenge of diagnosing and mitigating concurrency bugs. A suite of novel instrumentation schemes will be developed for monitoring thread interleaving patterns. Coupled with statistical debugging models developed previously, this lets developers identify bad thread interleavings which constitute root causes of program failure. A new approach to coordinated cross-thread random sampling keeps overheads low while still providing ample data for diagnosis. Static analysis will play a role to further reduce instrumentation load. Prior statistical debugging work was content with diagnosis only, but this project will develop a speculative locking strategy, guided by the statistical models, to avoid and thereby mitigate the effects of a variety of concurrency bugs.

计算机技术正在迅速渗透到社会的各个领域。一个影响数千或数百万人生活的计算机系统创造了一个庞大的用户社区，他们对该系统的正确行为感兴趣。广泛的互连意味着我们现在有能力挖掘这一潜力。这项工作面临着诊断和缓解并发错误的挑战。将开发一套新的插装方案来监控线程交织模式。再加上之前开发的统计调试模型，这使开发人员能够识别构成程序失败根本原因的不良线程交错。一种新的协调跨线程随机抽样方法保持了较低的开销，同时仍为诊断提供了充足的数据。静态分析将起到进一步降低仪器负荷的作用。以前的统计调试工作只满足于诊断，但这个项目将开发一种以统计模型为指导的推测性锁定策略，以避免并从而减轻各种并发错误的影响。