Bradley
Glasbergen,
PhD
candidate
David
R.
Cheriton
School
of
Computer
Science
Data systems continue to grow in complexity in response to the need to manage vast quantities of data and support a variety of workloads. Small changes in workloads and system configuration can result in different system behaviour and performance characteristics. As a result, DBAs and system developers can spend many hours diagnosing and debugging performance problems in data systems and the applications that use them.
I will present Sentinel, a system that constructs fine-grained models of system behaviour and enables diagnostic comparisons to understand how this behaviour differs for different workloads and system configurations. Sentinel’s insights are derived from built-in debug logs without necessitating that these logs are written to disk, thereby generalizing to all systems that use debug logging without incurring its overheads.