Analysis of Resource Consumption Within Consolidated Environments Using the Example of Standard Commercial Software

grrrgrapeInternet and Web Development

Oct 31, 2013 (3 years and 1 month ago)

101 views

Analysis of Resource Consumption
Within Consolidated Environments

Using the Example of Standard
Commercial Software

Daniel Richter

Master’s Thesis

24 November 2009

SAP Hewlett
-
Packard Competence
Center


located in SAP Partner Port,
Walldorf


partnership since 1989


nearly 50% of all SAP installations

worldwide runs on HP hardware



make SAP products run optimal on HP hardware


consultancy, best practices



Dr.
-
Ing
. Michael Mißbach (Senior Solution

Expert

and SAPS
-
meter product manager)


Master's Thesis | Daniel Richter | 24 November 2009

2

Outline


Basics: SAP System Landscape


Motivation, Related Work


SAP Application Server vs. Database Load Ratio


Performance Indicators, Base Data


Load Profiles Tools


Summary & Outlook




Master's Thesis | Daniel Richter | 24 November 2009

3

SAP System
Landscape
:

SAP System
Architecture


Presentation Layer


personal computer


Application Layer


SAP Application Server


Java, APAP


Data Layer


Database


business data


application code

Master's Thesis | Daniel Richter | 24 November 2009

4

SAP System
Landscape
:

System
Usage


Productive


real data


mission critical


Development


customizing


software lifecycle


Quality Assurance


copy of productive system


tests (functional, regression, stress)


Sandpit


demo, training


Master's Thesis | Daniel Richter | 24 November 2009

5

SAP System
Landscape
:

SAP Software Solution


Enterprise Core
Component

(
ECC
)
T


Business Warehouse (
BW
)

A


Customer
Relationship

Management (
CRM
)

T


Supplier

Relationship

Management (
SRM
)

T


Supply

Chain Management (
SCM
)

A


Exchange Infrastructure (
XI
)


Enterprise Portal (
EP
)


Solution Manger (
SolMgr
)


NetWeaver

Development Infrastructure (NWDI)


Global Trade System (GTS)




T = OLTP

A = OLAP


Master's Thesis | Daniel Richter | 24 November 2009

6

Motivation

Master's Thesis | Daniel Richter | 24 November 2009

7

Motivation

Master's Thesis | Daniel Richter | 24 November 2009

8

Motivation

Master's Thesis | Daniel Richter | 24 November 2009

9

Motivation

Master's Thesis | Daniel Richter | 24 November 2009

10

Motivation


Consolidation


Application Stacking


(hypervisor
-
based) virtualization



Sizing


too small: bad performance, crash


to big: resource wasting (energy, money, etc.)



Aim: predict resource consumption


SAP Application Server vs. database


SAP solution vs. SAP solution


Productive vs. Development vs. Quality Assurance

Master's Thesis | Daniel Richter | 24 November 2009

11

Related

Work


“Statistical Analysis of
ressource

demand of Standard software
solution”,
Meraj

Ahmed, 2003



Statistische

Analyse

von
Speichersystemen

in SAP
Umgebungen

und
Entwicklung

einer

Software
-
Anwendung

zur

optimalen

Dimensionierung

solcher

Systeme
”, Christoph Schelling, 2006



Analyse

des
Ressourcenverbrauchs

in
virtualisierten

Systemen

-

Am
Beispiel

der E/A
-
Last
betriebswirtschaftlicher

Standardanwendungen
“, Robert Wierschke, 2009


“MVS Performance Management Reporting”, Gary Hall and Linda
August, 1992


“Performance and Scalability Models for a
Hypergrowth

e
-
Commerce Web Site”, Neil J.
Gunther
, 2001


many research for
webserver


Master's Thesis | Daniel Richter | 24 November 2009

12

Performance
Indicators


CPU, IOPS, RAM


SAP
Application

Performance Standard (SAPS)



SAPS Meter

Master's Thesis | Daniel Richter | 24 November 2009

13

Base Data


14 companies (4x automobile, 3x manufacturing, 3x chemical, 1x
outsourcing, 1x retail, 1x logistics, 1x aerospace & defense


186 hosts


441 SAP systems


44 productive, 35 development, 40 quality assurance


45x ECC, 21x BW, 10x CRM, 8x ISA, 7x EP, 7x XI, 5x
SolMgr
,
4x SCM, 3x SRM, 3x LES, 3x GTS, 2x NWDI, 1x PMLS


64 Unicode, 39 no Unicode



3,732,341 raw measurements (CPU, #cores, IOPS, RAM)


14,093,748 aggregated measurements (SAPS, IOPS, RAM)


Master's Thesis | Daniel Richter | 24 November 2009

14

Database vs. SAP
Application

Server
Load

Ratio


SAP Application Server variable scalable


only one single database



previous assumptions:


SAPS


OLTP (ECC, CRM, SRM): 1:3


1:4 (DB/AS)


OLAP (BW, SCM): 1:1


1:2


EP, XI: sparse DB load


IOPS


negligible load of SAP Application Server


RAM


Master's Thesis | Daniel Richter | 24 November 2009

15

Database vs. SAP
Application

Server
Load

Ratio | SAPS


influcence

of SAP solution (SAPS)












influence probable


T = OLTP

A = OLAP







Master's Thesis | Daniel Richter | 24 November 2009

17

SAP

solution

load

ratio

(DB/AS)

sample

size

confidence

interval


(
database

part
)

SCM

A

69

%

/

31

%


1

sample

size

too

small

ECC

T

40

%

/

60

%


4

[
27
.
16

%
;
53
.
25

%
]

BW

A

29

%

/

71

%


12

[
5
.
77

%
;
53
.
12

%
]

CRM

T

26

%

/

74

%


1

sample

size

too

small

GTS


21

%

/

79

%


2

[
13
.
01

%
;
28
.
57

%
]

SolMgr


20

%

/

80

%


3

[
-
83
.
09

%
;
123
.
85

%
]

XI


20

%

/

80

%


2

[
-
27
.
52

%
;
67
.
08

%
]

Portal


9

%

/

91

%


7

[
4
.
37

%
;
13
.
44

%
]

Database vs. SAP
Application

Server
Load

Ratio | SAPS


influence of system usage (SAPS):








influence possible



influence of Unicode (SAPS):







influence probable (statistically firm)



Master's Thesis | Daniel Richter | 24 November 2009

18

usage

load

ratio

(DB/AS)

sample

size

confidence

interval

(
database

part
)

Development

22

%

/

78

%


11

[
10
.
65

%
;
33
.
15

%
]

Productive

29

%

/

71

%


12

[
13
.
21

%
;
45
.
41

%
]

Quality

Assurance

35

%

/

65

%


14

[
19
.
72

%
;
49
.
73

%
]


Unicode?

load

ratio

(DB/AS)

sample
size

confidence

interval

(
database

part
)

No

58

%

/

42

%


7


[
39
.
08

%
;
75
.
93

%
]

Yes

26

%

/

74

%


28


[
17
.
64

%
;
33
.
91

%
]

Database vs. SAP
Application

Server
Load

Ratio | SAPS


influence of SAP solution and Unicode (SAPS)








summary


statistical significant difference between usage of Unicode and
no usage of Unicode


influence depends on SAP solution

19

Unicode?

SAP
solution

load

ratio

(DB/AS)

sample

size

confidence

interval

(
database

part
)

No

BW

54 % / 46 %

3

[5.78 %;101.36 %]

No

ECC

58 % / 42 %

3

[
-
9.63 %;124.96 %]

Yes

BW

21 % / 79 %

1

sample
size

too

small

Yes

ECC

35 % / 65 %

10

[26.06 %;47.88 %]

Database vs. SAP
Application

Server
Load

Ratio | IOPS


influence of SAP solution (IOPS):












influence possible










Master's Thesis | Daniel Richter | 24 November 2009

20

SAP

solution

load

ratio


(DB/AS)

sample

size

confidence

interval

(
database

part
)

SCM


70

%

/

30

%


1

sample

size

too

small

ECC


69

%

/

31

%


13

[
50
.
28

%
;
88
.
37

%
]

BW


67

%

/

33

%


7

[
43
.
54

%
;
90
.
36

%
]

Portal


60

%

/

40

%


7

[
38
.
93

%
;
80
.
99

%
]

CRM


59

%

/

41

%


1

sample

size

too

small

GTS


44

%

/

56

%


3

[
-
0
.
34

%
;
88
.
43

%
]

Solution

Manager


42

%

/

58

%


2

[
-
339
.
3

%
;
423
.
19

%
]

XI


18

%

/

82

%


2

[
-
45
.
32

%
;
82

%
]

Database vs. SAP
Application

Server
Load

Ratio | IOPS


influence of system usage:








influence possible



influence of Unicode







influence probable (statistically firm)

Master's Thesis | Daniel Richter | 24 November 2009

21

usage

load

ratio


(DB/AS)

sample

size

confidence

interval

(
database

part
)

Development

50

%

/

50

%


11

[
29
.
93

%
;
69
.
45

%
]

Productive

70

%

/

30

%


12

[
54
.
24

%
;
85
.
05

%
]

Quality

Assurance

65

%

/

35

%


14

[
47
.
9

%
;
81
.
13

%
]


Unicode?

load

ratio

(
DB/AS)

sample

size

confidence

interval

(
database

part
)

No

80

%

/

20

%


7


[
68
.
41

%
;
91
.
6

%
]


Yes

54

%

/

46

%


28


[
42
.
94

%
;
65
.
77

%
]

Database vs. SAP Application Server
Load Ratio | Summary


SAPS

load


depends on SAP solution and Unicode


previous assumptions OK


except OLTP systems


here 1:1 to 1:2 (DB/AS) (as
OLAP)


IOPS

load:


IOPS of SAP
AppServer

not negligible!


depends on Unicode; SAP solution and usage possible (but not
statistically significant)


Master's Thesis | Daniel Richter | 24 November 2009

22

Load

Profiles


predict resource usage at specific point in time


corresponding to power supply industry


in the curse of the day


separated by weekday, Saturday, Sunday


scatter view







this analysis: only SAPS


corresponding IOPS mostly similar


Master's Thesis | Daniel Richter | 24 November 2009

23

Load

Profiles


productive OLTP systems (online transactional processes)

Master's Thesis | Daniel Richter | 24 November 2009

24

Mon
-
Fri

Sat

Sun

ECC

ECC

ECC

ECC

Load

Profiles


productive OLAP systems (online analytical processes)

Master's Thesis | Daniel Richter | 24 November 2009

25

Mon
-
Fri

Sat

Sun

BW

BW

BW

SCM

Load

Profiles


other productive systems

Master's Thesis | Daniel Richter | 24 November 2009

26

Mon
-
Fri

Sat

Sun

EP

EP

XI

XI

Load

Profiles

Master's Thesis | Daniel Richter | 24 November 2009

27


SAP solution load ratio

customer

K
1

manufact
.

K
2

chemical

K
3

outsourc
.

K
4

automot
.

K
5

manufact
.

avg
.

ECC

42
%

45
%

99
%

70
%

42
%

59
%

BW

58
%

7
%

0
%

10
%

14
%

18
%

SolMgr

0
%

23
%

0
%

0
%

1
%

5
%

XI

0
%

6
%

0
%

0
,
6
%

16
%

4
%

SCM

0
%

0
%

0
%

0
%

22
%

4
%

Portal

0
%

2
%

1
%

14
%

0
%

3
%

GTS

0
%

11
%

0
%

0
%

0
%

2
%

NWDI

0
%

7
%

0
%

0
%

0
,
1
%

1
,
5
%

SRM

0
%

0
%

0
%

5
%

0
%

1
,
1
%

CRM

0
%

0
%

0
%

0
%

3
%

0
,
7
%

LES

0
%

0
%

0
%

0
%

1
%

0
,
3
%

PMLS

0
%

0
%

0
%

0
%

0
,
9
%

0
,
2
%

ISA

0
%

0
%

0
%

0
%

0
,
3
%

0
,
1
%

Load

Profiles


Productive vs. Development vs. Quality Assurance


individual for every SAP system


load ration for different customers:

Master's Thesis | Daniel Richter | 24 November 2009

28

cust
.

industry

productive

development

quality

assurance

K
1

manufacturing

88
%

8
%

4
%

K
2

chemical

35
%

41
%

24
%

K
3

outsourcing

84
%

5
%

10
%

K
4

automotive

73
%

2
%

25
%

K
5

manufacturing

68
%

18
%

14
%

Average

70
%

15
%

16
%

Load

Profiles


base load


mostly negligible


average load


OLTP well predictable


peak load


long
-
term


periodic, short
-
term

Master's Thesis | Daniel Richter | 24 November 2009

29





Tools


data download tool


website & GUI controls












https://sapsmetering.asg
-
platform.org:8080/


Master's Thesis | Daniel Richter | 24 November 2009

30

Summary

& Outlook


examined key factors for DB vs. SAP
AppServer

load ratio


SAP solution, system usage


Unicode


confirmed and corrected previous assumptions


load profile overview for SAP solutions


OLTP good predictable


OLAP not predictable



more, detailed data is needed

Master's Thesis | Daniel Richter | 24 November 2009

31