Improving distributed system performance using machine availability prediction
James W. Mickens, Brian D. Noble · ACM SIGMETRICS Performance Evaluation Review · 2006
In a distributed system, a set of networked machines provides a highly available service to remote clients. Traditional distributed systems like AFS [2] make a clear distinction between clients and servers. Client machines may be poorly administered, cheaply constructed, often offline, and possibly malicious. In contrast, servers are expected to be well-administered and almost always online. Highly available servers ensure the availability and reliability of the distributed service.