Carrying the crash-only software concept to the legacy application servers

Abstract

In the last few years, high-availability on internet services has become a main goal for the academia and industry. We all know how complex and heterogeneous Internet service systems are and how sensitive to suffer from transient failures or even crashes also. Because developing systems that are guaranteed to never crash and never suffer transient or intermittent failures seems an impractical and unfeasible business, there is a need to develop mechanisms that can suffer crashes and transient failures as if they were a clean shutdown. Behind this idea, the creators of the crash-only software concept proposed a new design strategy in order to get crash-safe and fast recovery systems by defining a list of laws which are needed in order to achieve that goal. However, their proposals are focused on new systems design. For this reason, we will discuss how to develop crash-safe and masked fast self-recovery legacy systems following the ideas behind the crash-only software concept. In our work, we have focused on legacy application servers because they are a more sensitive piece of the internet services' big puzzle.