Speeding up profiling program's runtime characteristics for workload consolidation

Area of Science:

• Computer Science
• Systems and Networking
• Performance Engineering

Background:

• Workload consolidation enhances resource utilization in data centers.
• Effective consolidation requires understanding program runtime characteristics, specifically resource sensitivity.
• Current methods for measuring resource sensitivity are time-consuming.

Purpose of the Study:

• To propose a fast and accurate method for profiling program resource sensitivity.
• To enable efficient workload consolidation by providing detailed runtime characteristics.
• To reduce the time and experimental overhead associated with resource sensitivity analysis.

Main Methods:

• Developed a two-level profiling acceleration strategy.
• Utilized resource usage information to set upper bounds for resource control, narrowing the experimental range.
• Employed an interpolation-based prediction model to estimate resource sensitivity, reducing direct program executions.

Main Results:

• The proposed method significantly shortens profiling time.
• The two-level strategy accurately captures program resource sensitivity.
• Average absolute error for resource sensitivity was controlled within 0.05.

Conclusions:

• The fast profiling method effectively accelerates resource sensitivity analysis.
• This technique supports more efficient and accurate workload consolidation in data centers.
• The approach balances speed and accuracy in performance profiling.