Uppsala Architecture Research Team
Our measurement technologies focus on gathering low-overhead data from full-scale applications on real hardware. This allows us to investigate the behavior of long-running applications and commodity systems. We capture both architecturally-independent data (e.g., StatCache Statistical Memory Modeling) as well as architecture-specific measurements (e.g., Cache Pirate).
Shared Resource Sensitivity
By stealing shared resources with a Pirate/Bandit application, we can quantitatively measure the resulting performance impact on a target application. This technique allows us to capture quantitative data on shared resource sensitivity on commodity hardware for full applications.
Online Phase Detection
Our low-overhead (2%) online phase detection tool (ScarPhase) allows us to accurately and quickly understand the detailed phase-behavior of a running application. With this information we can improve the efficiency and accuracy of memory modeling, power modeling, and resource contention modeling.