Computer systems often fail in production environment.  As these failures directly affect the customers, 
large computer system vendors typically have to invest significant amounts of resources in diagnosing them.  
Unfortunately, diagnosing these production failures is notoriously difficult.  Indeed, constrained by both 
privacy and expense reasons, software vendors often cannot reproduce such failures.  Therefore, support engineers 
and developers continue to rely on the logs printed by the run time system to diagnose the production failures.  
However, the ad-hoc natures of today's system logs are frequently insufficient for effective failure diagnosis.

In this talk, I will describe our work on improving the software logging for better production failure diagnosis.  
One approach, LogEnhancer, uses a novel combination of program analysis and system techniques to collect additional 
information for each existing log message.  Another approach, LogError, tackles the problem of 'silent failures' --- 
failures without any log messages printed.  We applied LogEnhancer and LogError to a broad range of real software systems, 
and found that we can significantly improve the postmortem failure diagnosis by improved software logging. 
The insights we learnt could also benefit programmers towards better designs of their software for better failure diagnosability.