Automatic on-line failure diagnosis at the end-user site

摘要: Production run software failures cause endless grief to end-users, and challenges programmers as they commonly have incomplete information about the bug, facing great hurdles reproduce it. Users are often unable or unwilling provide diagnostic due technical privacy concerns; even if is available, failure analysis time-consuming. We propose performing initial diagnosis automatically at end user's site. The moment of a valuable commodity strive reproduce-- leveraging it directly reduces effort while simultaneously addressing concerns. Additionally, we protocol. So far know, this first such automatic protocol proposed for on-line diagnosis. By mimicking steps human programmer follows dissecting failure, deduce important information. Beyond use, can also reduce in-house testing. We implement some these ideas. Using lightweight checkpoint rollback techniques dynamic, run-time tools, initiate several bugs. Our preliminary results show that efficiently accurately find likely root causes fault propagation chains. Further, normal execution overhead only 2%.