Upload
dannyq
View
462
Download
0
Embed Size (px)
DESCRIPTION
Now more than ever there is a need for organisations to ensure there is optimal infrastructure capacity in place to support business services. Excess capacity can result in unacceptable capital and operating costs, impacting business profitability. Conversely, insufficient capacity can impact service performance and business competitiveness. The Capacity Plan should determine the optimal capacity required and a key input into it is forecast service demand. This presentation details a number of techniques to forecast service demand using a business-driven approach. A number of important considerations are addressed, including business seasonality, forecast error and techniques for translating business demand to service and component demand. The techniques are demonstrated with case studies based on real-life client engagements.
Citation preview
itSMF 2009 Annual Conference
How to Deliver Business-Driven Demand Planning
Danny Quilton, COO, Capacitas
Agenda
� Overview� Business-driven demand planning � Challenges associated with business demand planning� Demand forecasting techniques� Demand management
Benefits of business-driven demand planning� Benefits of business-driven demand planning
2
Demand Planning Overview
� A key input into the Capacity Management process is the anticipated level of demand expected of the system
� Demand planning can be carried out at different ‘layers’� These layers are defined by ITIL:
Business
� Note that these layers apply to a single information and communication technology (ICT) service
Business
Service
Component
3
Demand Planning Overview
• Understood by the business
• May be forecast by the businessBusiness demand
• Functionality presented to the user• Functionality presented to the user
• May not be understood by the business
• Technology independent
• The link between business and component demand
Service demand
• Not understood by the business
• Technology specific; “bits and bytes”
• The actual consumer of capacity
Component demand
4
Demand Planning – Online Banking
Number of accounts
Make transfer
Server CPU demand
Check balance
Server CPU demand
Server memory demand
Show statement
Server I/O demand
Server CPU demand
5
Demand Planning – Corporate Messaging
Service
Number of users
Send Receive Create
Send emails
Server CPU
demand
Network demand
Receive emails
Server CPU
demand
Network demand
Create journal entries
Server I/O demand
Server CPU
demand
6
Demand Planning – e-commerce Service
Product Inventory
Add to Search
Server CPU
demand
Network demand
Add to Basket
Server CPU
demand
Network demand
Checkout
Server I/O demand
Server CPU
demand
7
Demand Planning – Mobile Phone Pre-Pay
Service
Number of Pay-as-you-go subscribers
Number of calls
Server CPU
demand
Server memory demand
Number of SMS
Server CPU
demand
Serer memory demand
Number of top ups
Server I/O demand
Server CPU
demand
8
Common Pitfall
40
50
60
70
80
90
100
Database Server Load; December 2007 - November 2008
0
10
20
30
40
20
07
-12
-02
12
:00
:00
AM
20
07
-12
-09
12
:00
:00
AM
20
07
-12
-16
12
:00
:00
AM
20
07
-12
-23
12
:00
:00
AM
20
07
-12
-30
12
:00
:00
AM
20
08
-01
-07
12
:00
:00
AM
20
08
-01
-14
12
:00
:00
AM
20
08
-01
-21
12
:00
:00
AM
20
08
-01
-28
12
:00
:00
AM
20
08
-02
-05
12
:00
:00
AM
20
08
-02
-12
12
:00
:00
AM
20
08
-02
-19
12
:00
:00
AM
20
08
-02
-26
12
:00
:00
AM
20
08
-03
-05
12
:00
:00
AM
20
08
-03
-13
12
:00
:00
AM
20
08
-03
-20
12
:00
:00
AM
20
08
-03
-27
12
:00
:00
AM
20
08
-04
-04
12
:00
:00
AM
20
08
-04
-11
12
:00
:00
AM
20
08
-04
-18
12
:00
:00
AM
20
08
-04
-25
12
:00
:00
AM
20
08
-05
-03
12
:00
:00
AM
20
08
-05
-10
12
:00
:00
AM
20
08
-05
-17
12
:00
:00
AM
20
08
-05
-24
12
:00
:00
AM
20
08
-05
-31
12
:00
:00
AM
20
08
-06
-08
12
:00
:00
AM
20
08
-06
-15
12
:00
:00
AM
20
08
-06
-22
12
:00
:00
AM
20
08
-06
-29
12
:00
:00
AM
20
08
-07
-07
12
:00
:00
AM
20
08
-07
-14
12
:00
:00
AM
20
08
-07
-21
12
:00
:00
AM
20
08
-07
-28
12
:00
:00
AM
20
08
-08
-05
12
:00
:00
AM
20
08
-08
-12
12
:00
:00
AM
20
08
-08
-20
12
:00
:00
AM
20
08
-08
-27
12
:00
:00
AM
20
08
-09
-04
12
:00
:00
AM
20
08
-09
-11
12
:00
:00
AM
20
08
-09
-18
12
:00
:00
AM
20
08
-09
-25
12
:00
:00
AM
20
08
-10
-03
12
:00
:00
AM
20
08
-10
-10
12
:00
:00
AM
20
08
-10
-17
12
:00
:00
AM
20
08
-10
-24
12
:00
:00
AM
20
08
-10
-31
12
:00
:00
AM
20
08
-11
-07
12
:00
:00
AM
20
08
-11
-14
12
:00
:00
AM
20
08
-11
-21
12
:00
:00
AM
20
08
-11
-29
12
:00
:00
AM
Database Server - CPU Utilisation - Max (%) Linear (Database Server - CPU Utilisation - Max (%))
9
Business-Driven Demand Planning
� Demand deconstruction
Business demand
Component demand
Service demand
10
Demand Deconstruction: Business to Service
� One unit of business demand will often map to many units of service demand
� Build an empirical understanding the relationships
Business demand
relationships� Consider the relationship
over the peak period
Component demand
Service demand
11
Challenges Planning Service Demand
Number of accounts
Make transfer
Check balance
Print statement
Change Password
Change address
Order credit card
Pay credit card bill
� Rich functionality – which service demand do I focus on?
� Poor instrumentation of the service
12
Demand Deconstruction: Business to Service
� Consider the peak rate of check balance
� Consider using segmentation:– Different account types will use the
system differently– E.g. Retail and Business accounts
Number of accounts
Check balance – E.g. Retail and Business accounts
13
balance
Demand Deconstruction: Service to
Component� A unit of service demand will be
implemented by one or more technical transactions
� The component capacity planner must identify these technical transactionsA technical transaction will
Business demand
� A technical transaction will traverse a number of components (infrastructure components)
� Each component in the path will be subjected to some component demand
14
Component demand
Service demand
Business Demand Planning
� Business demand is termed ‘business volume indicators (BVIs)
Identify business stakeholders
Agree suitable BVIs
Measure BVIs
Forecast BVIs
15
Criteria for Defining BVIs
BVIs must be understood by the business
BVIs must have a direct bearing on system capacity
Selected BVIs must have ‘buy in’ from the business
BVIs must be measurable
16
Example BVIs from Client Engagements
Airline
Aircraft
Internet Bank
Broadband Service
Provider
Subscribers
Retailer
Stock keeping
units (SKUs)
Stock Broker
Trading staff
Broadcaster
Aircraft
Airports
Number of accounts
Subscribers
Exchanges
units (SKUs)
Stores
Lorry Deliveries
Trading staff
Trades
Subscribers
17
Tips for Measuring BVIs
� It is essential that BVIs are measured in production� BVIs cannot be forecast if current BVI levels are unknown� Sources of BVI data:
– Database systems are likely to hold BVI information– Application monitors– Audit logs – Audit logs
� BVIs are typically measured at coarse sample intervals, e.g:– Monthly– Quarterly
� Service acceptance process must demand BVI monitoring
18
Business Forecasting Challenges
Lack of engagement from the business
Business demand is confidential or
commercially sensitive
Over optimistic forecasts from the business
Lack of forecasting skills within the business
19
Establishing Business Demand Forecasts
� The preference is always to work with the business to establish a business demand forecast
� There will however be occasions where business demand forecasts are not forthcoming
� Then the capacity management function will need to establish a � Then the capacity management function will need to establish a business demand forecast
20
Sources of Business Demand Forecasts
� Sales and marketing revenue forecasts� HR headcount projections� Business cases for new services� Research from external bodies, e.g:
– Ofcom http://www.ofcom.org.uk/research/telecoms/reports/– Office for National Statistics http://www.statistics.gov.uk/– Office for National Statistics http://www.statistics.gov.uk/– Research companies (Gartner, Ovum, Forrester, etc.)– Competitors (via annual company reports)
21
Forecasting Techniques
� Linear trending � Time series decomposition� Forecast error
22
Linear Trend Forecast for an Internet Banking
Service
y = 7438.3x
R² = 0.9766
400,000
500,000
600,000
700,000
Re
gis
tere
d U
sers
Historical Business Demand Since Go-live
0
100,000
200,000
300,000
Jan
-20
02
Ma
r-2
00
2
May
-20
02
Jul-
20
02
Se
p-2
00
2
No
v-2
00
2
Jan
-20
03
Ma
r-2
00
3
May
-20
03
Jul-
20
03
Se
p-2
00
3
No
v-2
00
3
Jan
-20
04
Ma
r-2
00
4
May
-20
04
Jul-
20
04
Oc
t-2
00
4
De
c-2
00
4
Fe
b-2
00
5
Ap
r-2
00
5
Jun
-20
05
Au
g-2
00
5
Oc
t-2
00
5
De
c-2
00
5
Fe
b-2
00
6
Ap
r-2
00
6
Jun
-20
06
Au
g-2
00
6
Oc
t-2
00
6
De
c-2
00
6
Fe
b-2
00
7
Ap
r-2
00
7
Jun
-20
07
Au
g-2
00
7
De
c-2
00
7
Fe
b-2
00
8
May
-20
08
Jul-
20
08
No
v-2
00
8
Re
gis
tere
d U
sers
23
Linear Trend Forecast for an Internet Banking
Service
y = 10538x + 312801
R² = 0.9947
400,000
500,000
600,000
700,000
Re
gis
tere
d U
sers
Historical Business Demand - 36 months to Dec 2008
0
100,000
200,000
300,000
Jan
-20
06
Fe
b-2
00
6
Ma
r-2
00
6
Ap
r-2
00
6
Ma
y-2
00
6
Jun
-20
06
Jul-
20
06
Au
g-2
00
6
Se
p-2
00
6
Oc
t-2
00
6
No
v-2
00
6
De
c-2
00
6
Jan
-20
07
Fe
b-2
00
7
Ma
r-2
00
7
Ap
r-2
00
7
Jun
-20
07
Jun
-20
07
Jul-
20
07
Au
g-2
00
7
Oc
t-2
00
7
De
c-2
00
7
Jan
-20
08
Fe
b-2
00
8
Ma
r-2
00
8
Ma
y-2
00
8
Jun
-20
08
Jul-
20
08
Se
p-2
00
8
No
v-2
00
8
De
c-2
00
8
Re
gis
tere
d U
sers
24
Linear Trend Forecast for an Internet Banking
Service
600,000
800,000
1,000,000
1,200,000R
eg
iste
rd U
sers
Forecast Business Demand
0
200,000
400,000
Jan
-20
02
Ap
r-2
00
2
Jul-
20
02
Oc
t-2
00
2
Jan
-20
03
Ap
r-2
00
3
Jul-
20
03
Oc
t-2
00
3
Jan
-20
04
Ap
r-2
00
4
Jul-
20
04
No
v-2
00
4
Feb
-20
05
May
-20
05
Au
g-2
00
5
No
v-2
00
5
Feb
-20
06
May
-20
06
Au
g-2
00
6
No
v-2
00
6
Feb
-20
07
Jun
-20
07
Oc
t-2
00
7
Feb
-20
08
May
-20
08
Sep
-20
08
Jan
-20
09
Ap
r-2
00
9
Jul-
20
09
Oc
t-2
00
9
Jan
-20
10
Ap
r-2
01
0
Jul-
20
10
Oc
t-2
01
0
Jan
-20
11
Ap
r-2
01
1
Jul-
20
11
Oc
t-2
01
1
Re
gis
terd
Use
rs
Historical Registered Users Forecast Registered Users
25
Business Demand of www.easyJet.com
y = 0.0461x - 1673.1R² = 0.9668
100
120
140
160
180
Number of Owned A
ircraft
Fleet Plan to April 2009
GB Airways
acquisition
0
20
40
60
80
Mar-04
Jun-04
Sep-04
Dec-04
Mar-05
Jun-05
Sep-05
Dec-05
Mar-06
Jun-06
Sep-06
Dec-06
Mar-07
Jun-07
Sep-07
Dec-07
Mar-08
Jun-08
Sep-08
Dec-08
Mar-09
Number of Owned A
ircraft
Delivery Date
26
Demand Seasonality
Dai
ly P
urc
ha
ses
Historical Service Demand for an e-commerce Service
y = 12.088x - 436266
R² = 0.9436
26/03/2004
26/04/2004
26/05/2004
26/06/2004
26/07/2004
26/08/2004
26/09/2004
26/10/2004
26/11/2004
26/12/2004
26/01/2005
26/02/2005
26/03/2005
26/04/2005
26/05/2005
26/06/2005
26/07/2005
26/08/2005
26/09/2005
26/10/2005
26/11/2005
26/12/2005
26/01/2006
26/02/2006
26/03/2006
26/04/2006
26/05/2006
26/06/2006
26/07/2006
26/08/2006
26/09/2006
26/10/2006
26/11/2006
26/12/2006
26/01/2007
26/02/2007
26/03/2007
26/04/2007
26/05/2007
26/06/2007
26/07/2007
26/08/2007
26/09/2007
26/10/2007
26/11/2007
26/12/2007
26/01/2008
26/02/2008
26/03/2008
26/04/2008
26/05/2008
26/06/2008
26/07/2008
26/08/2008
26/09/2008
26/10/2008
26/11/2008
26/12/2008
26/01/2009
26/02/2009
26/03/2009
26/04/2009
26/05/2009
26/06/2009
Dai
ly P
urc
ha
ses
Actual Daily Purchases Trend 180day Linear (Trend 180day)
27
Time Series Decomposition
Forecast Service Demand for an e-commerce Service
01
/01
/20
06
01
/04
/20
06
01
/07
/20
06
01
/10
/20
06
01
/01
/20
07
01
/04
/20
07
01
/07
/20
07
01
/10
/20
07
01
/01
/20
08
01
/04
/20
08
01
/07
/20
08
01
/10
/20
08
01
/01
/20
09
01
/04
/20
09
01
/07
/20
09
01
/10
/20
09
01
/01
/20
10
01
/04
/20
10
01
/07
/20
10
01
/10
/20
10
01
/01
/20
11
01
/04
/20
11
01
/07
/20
11
01
/10
/20
11
01
/01
/20
12
Actual Daily Purchases Forecast Daily Purchases
28
Forecast Error
� Any forecast you make will be wrong!
� The key step is to measure your to measure your forecast error
29
Forecast Error
� Forecast error is the difference between what was forecast and what actually occurred
� Forecast error, et is given by:
ttt FAe −=
– At is the observed value at time period t– Ft is the forecast value at time period t
30
Forecast Error
t
ttt
A
FAPE
−
=� Percentage error:
Mean percentage error:
PE
MPE
n
1t
t∑=
=
� Mean percentage error:
� Mean absolute percentage error:
nMPE 1t=
=
n
|PE|
MAPE
n
1t
t∑=
=
31
Forecast Error
10,000
12,000
14,000
16,000
Pu
rch
ase
s
Forecast Service Demand vs. Actual Service Demand
0
2,000
4,000
6,000
8,000
Oct-07 Nov-07 Dec-07 Jan-08 Feb-08 Mar-08 Apr-08 May-08 Jun-08 Jul-08 Aug-08 Sep-08 Oct-08 Nov-08 Dec-08
Pu
rch
ase
s
Actual Bookings (Peak Hour) Forecast Bookings (Peak hour)
32
Forecast Error
�Here the MAPE is 12%
Aug-08
Sep-08
Oct-08
Nov-08
Dec-08
Forecast Error
-20% -15% -10% -5% 0% 5% 10% 15% 20%
Oct-07
Nov-07
Dec-07
Jan-08
Feb-08
Mar-08
Apr-08
May-08
Jun-08
Jul-08
33
Other Forecasting Techniques
� Moving average smoothing methods� Exponential smoothing methods
34
Extraordinary Peak Demand
Service Extraordinary Peak Scenario
Internet banking service Run on a bank
News service Major news event, e.g. 9/11News service Major news event, e.g. 9/11
Mobile
telecommunications
Major news event, e.g. 7/7
New Years Eve
E-commerce service Unexpected demand resulting from a
promotion
35
Demand Management
36
Demand Management
37
Benefits of Business-Driven Demand Planning
Demand forecasts can be signed off by the business
Capacity plans can be driven directly by business volumes
Business-driven
Justification for capacity upgrades
Justification for SLA modifications
Business-driven demand planning
38
Summary
� Business driven capacity planning requires planning activities at all 3 ITIL layers:– Business – Service – Component
� Business demand drives the Capacity Management process� Business demand drives the Capacity Management process– Must have business ‘buy in’
� Component demand dictates the capacity requirements� Service demand provides the translation between business
and component demand� Demand deconstruction� This approach may be warranted for your important ICT
services only
39
Questions?
� Please visit us at our stand at P09 for any further
questions
� Presentation will be available for download from
www.capacitas.co.ukwww.capacitas.co.uk
40