3. Predictive Modelling Need
• Insights into IT systems behavior (non-functional) ahead
• Better predictability of
➢ System responsiveness
➢ Infrastructure utilizations
• Ability to plan for hardware in advance for Mergers &
Acquisitions / Country Rollouts / New Features etc.
• Minimize surprises w.r.t. performance & scalability
4. Predictive Performance Modeling
4
What? Predict Quality of Service (QoS) of IT systems (response time, throughput,
server utilization) for varying conditions such as increase in # of users /
transaction load / hardware configurations - by making use of available data
points from Production / QA and forecast
How it helps?
Data-driven decision making for additional hardware procurement for future workloads
Effective hardware utilization
Provides leads for quicker tuning / optimization of specific layers of Application
Landscape
Reduces cost & time to carry out multiple trial-error performance runs
KEY CONSIDERATIONS
Not a replacement for performance benchmarking /
Application tuning – complements them
More the data - more accurate the predictions
Analytical models are cost effective – can be
implemented in multiple ways
Thumb
Rules
Linear
Projections
Analytical
Models
Simulation
Techniques
Real
Environment
Accuracy
Cost
Performance
Prediction
Techniques
QNM
ML
Empirical
Formula
5. Key Components ofMachine Learning for Performance Modeling
5
Trained Data Set Test Data Set
ML-based Performance
models
Prediction results