methods of forecasting for capacity management
TRANSCRIPT
Abstract
Forecasting is the process of making statements about events in the future. Events related to capacity management are typically things like the state of resource consumption, service levels, and computing environment changes at future points in time. Making statements or predictions about these future events requires analysis of information to determine a future state. Knowing what information is needed to make accurate forecasts is a critical step for any analysis.
Forecasts are made to answer questions. Understanding the questions, and things that affect answers to those questions, is the first step to creating an accurate forecast. Required accuracy of a forecast should determine which methods are used to create it. Assumptions can be made to limit the amount of data and time required for creating forecasts. Validating forecast accuracy, after events happen, is an important part of continually improving future forecasts, and building credibility. This paper describes the important task of forecasting as it relates to capacity management.
2
Agenda
Why do we forecast?
Forecasting scenarios
Forecasting Techniques
Forecasting and Virtualization
Summary
3
Why do we forecast?
What are we trying to accomplish?
Forecasting Objectives
– Make business decisions
– Resolve possible conflicts
– Provide Efficiency
– Balance Cost vs. Service
5
Forecasting
Forecasting relies on
– Proper technical and business data
– Proper business information
– Valid technical and business assumptions
– Ability to compare past activity
– Match needs and cost
7
Forecasting Scenarios
Types of scenarios
– Near term – 3 to 12 months
– Long term – 1 to 3 years
– Environment changes
• Down sizing Infrastructure
• Right sizing Infrastructure
• IT mergers and acquisitions
8
Forecasting – Where do we get the data?
Data collected from various tools
Past forecasts
Business users and owners
– May want to help
– May say “You’re the capacity team, you figure it out”
9
Forecasting Techniques
11
Mo
re A
ccu
rate
More Effort
Benchmarking
SimulationMeasurement
And Analytical
Modeling
Trending
Charts
Trending
Provide prediction based upon an interval of time
Quick to produce
Usually looking at one metric
Can show steps in growth changes
Confidence is based on length of interval and amount of historical data
13
Analytical Modeling
The ability to predict environment changes based upon the arrival of work
– CPU
– IO
Based upon
– Baseline timeframe
– Calibration of model – Does it match real life?
– What-If changes to model
Spend virtual $’s
Low cost modeling technique
18
Simulation Modeling
Replication of environment
Has to stay close to production environment
Usage of tools to replicate workload into simulated environment
Longer lead time to setup
Provides granular detail on environment changes
Can be costly
21
Forecasting and Virtualization
Compact environment into smaller footprint
Reducing IT costs
How much more can I fit, i.e. how much headroom do I have?
22
The Correct Technique?
What is the end-user expecting?
Which technique gives you the best result for the question asked?
How much time do you have to produce a result?
23
Summary
Understand what the forecast is to accomplish
What do enterprise standards dictate?
Ensure you have as much business and technical information as possible
Trend when trends make sense
Repeatable process
Document process and assumptions
24