Vous êtes sur la page 1sur 15

Capacity

Management at
Dominos Pizza Group
Danny Quilton, Capacitas
Gavin Greenfield, Dominos Pizza
Group

Agenda

The problem
The approach
The outcome
Q&A

The Problem Domain

Some background

Dominos ecommerce platform is an increasingly important element of


the commercial strategy
Dominos had a five year business and technology plan ending 2011
Five year plan coincides with Hardware refresh
Strategy for new 5 year plan started end of 2010
Strategy Implementation to be completed Q4 2011
Including new corporate communications network
Reorganisation of IT department
Outsource of server & infrastructure to a hosted environment

Objectives

Look at the performance of existing Infrastructure


Confirming or refuting the feelings of where our bottlenecks were or
might next occur
Options to ensure that the existing environment will perform
acceptably until decommissioned
Sizing information for new environment
Potential sizing requirements into 2012/2013
Options on scaling the new environment when required

Technical Scope

Scoping exercise carried out to identify high-risk components:

Public internet

Network
access points

Credit card
authorisation thirdparty

Content delivery
network

Firewall platform

Load balancer
platform

Web
platform

Database
platform

The Approach

Demand
modelling

Capacity and
performance
modelling

Capacity Plan

Capacity Planning Process

Demand Modelling
Business Demand

Number of Dominos stores

Service Demand

Number of e-commerce orders

Component Demand

Database locking demand

Server CPU load


Network load
Firewall CPU load

Business Demand Forecasting

Business strategy to increase the number of stores

10

Service Demand

Service consists of a 5-step booking funnel:

11

Service Demand

Analysis of historical demand identified strong seasonality:

12

Service Demand

Note the underlying growth:


Mobile application adoption
Increasing stores

Seasonality identified using


statistical techniques:
Friday
Saturday
Two for Tuesday promotion
Winter

13

Service Demand

e-commerce
services
typically exhibit
high peak to
average
demand

14

Service Demand Forecast

A forecast using
the forecast
growth in stores
and historical
seasonality

15

Capacity Modelling
Component
capacity
loading is
driven by
the number
of orders
Capacity
required is
forecast
using the
model and
forecast
peak hour
orders

16

Capacity Modelling

Component
capacity
loading is
driven by
the number
of orders

17

Capacity Modelling

Akamai
supplies a
content
delivery
network.
Operating cost
is related to
traffic carried.

18

Capacity Modelling

Firewall has an Intrusion Prevention Service module with onboard


CPU
CPU capacity needed to be planned
When IPS capacity is reached the response time the user experiences
when using the service increases

19

Capacity Modelling

Conclusion?

The relationship
changes from
09/02/2011
onwards

20

10

Capacity Modelling

Modelled relationship over whole period:

21

Capacity Modelling

Modelled relationship after 09/02/2011:

22

11

Capacity Modelling

Database CPU utilisation was not driven by the volume of orders


Instead:
Number of concurrent users
Rate of database lock requests
23

Capacity Modelling

Evidence of potential
bottleneck with regard to
database disk I/O

Evidence of potential
bottleneck with regard to
database locking

24

12

Capacity Planning Scenarios


Business Capacity
Management

Forecast number of stores in 2012 is X

Service Capacity
Management

The forecast peak hour number of ecommerce orders is Y

Component Capacity
Management

The firewall IPS CPU would be 125%


loaded

25

The outcome

Clear identification of required capacity over next 3-years


Delivery of a repeatable process
Detailed analysis of the traffic flow over the firewall lead to a look at
the volume of traffic to the database server
Moving of session state services to different network segment
Long term revision to use of session state

Identification and rewriting of the inefficient queries that were leading


to the database locking
Reducing both load and I/O

More accurate forecast of firewall requirements for new hosted


environment

26

13

Next Steps
Regular capacity plan updates
Demand forecast validation
Component capacity forecast validation on an on-going basis

27

Summary

A business-driven capacity planning approach


Demand planning followed by capacity modelling
Significant risk mitigation achieved
This approach is applicable to any IT service

28

14

Questions?

Please visit us at the Capacitas stand


www.capacitas.co.uk

29

15

Vous aimerez peut-être aussi