Architecture and Database Interfaces

Donald Kossmann

Outline
1. SAP's Client-server Architecture
2. Overview of the SAP Schema
3. Transaction Processing
4. Benchmarks and Results

SAP's Three-Tier Client-Server Architecture

Presentation

...
presentation servers: control GUI
LAN or WAN
application servers: carry out ABAP
appl. appl. programs and DynPros
server ... server
LAN RDBMS: stores all the data (including
ABAP programs, DynPros, data dic-
RDBMS tionary, etc.)

3
 Advantages of Three-Tier Architectures
1. Scalability. Add machines in middle tier to support more users
2. Portability. It is possible to use dierent platforms at all levels.
3. Interoperability and openness. Middleware serves as
platform to integrate and interact with third-party products.
4. Nice GUIs. Presentation servers can interact with Microsoft
Word, Excel, etc.
4
Integrating WWW Services

Transaction
Server

Internet
Server

server
Web

appl.
Internet

RDBMS
LAN

LAN
server. . .
Internet

local clients
clients

appl.

5
 SAP R/3 Congurations
tiny: 1 user
 all three layers on one machine
 one of the SAP founder's golf club is run by R/3 on a laptop
small: about 10 users
 PCs for presentation
 application and database server on one (mid-range) machine
 Ethernet

Serious SAP R/3 Congurations

medium: about 100 users
 PCs, notebooks, other workstations for presentation
 a couple of machines for application servers;
 one (fairly big) database server machine;
 Ethernet for local PCs;
big: more than 1000 users
 PCs, notebooks, other workstations for presentation;
 several machines for application servers;
 a mainframe/multi-processor machine for the database
 FDDI

7
 8
Supported Platforms
Presentation Layer
 Windows 3.1, Windows 95, Windows NT
 Java
 OSF/Motif
 OS/2
 Macintosh
 in the long run, only Windows and Java
sensors) are also very common

are going to be supported

Operating Systems for Application

Servers
Additional Gimmicks

 AIX, Digital Unix, HP-UX, SINIX,

 standby database server machine is recommended

SOLARIS
 usually, installations have separate machines for tests

 Windows NT
 OS/100 (for IBM AS/400)
 ISDN, DATEX-P, and special links to technical subsystems (e.g.,

9
  DB2 Common Server, DB2 for AS/400,

 Bull, IBM, SNI, SUN, Digital, HP for

(ADABAS D: only support for old

 LU 6.2 (for IBM mainframes)

 many, many for Windows NT

Communication Protocols
Database Systems

DB2 for OS/390

UNIX platforms
Informix Online

MS SQL Server

installations)

 IBM AS/400
 IBM S/390

 TCP/IP
Hardware
Oracle





10
Overview of the SAP Schema
 R/3 has more than 10,000 pre-dened tables (Version 3.x)
{ tables for data such as customers, orders, etc.
{ tables for statistics (monitoring the system)
{ tables for authorization
{ ...
{ comprehensive, generic schema for any kind of conceivable
business rather than greatest common denominator
 fully normalized, (almost) no redundancy
{ good for OLTP
{ bad for OLAP (as we will see)
 users can also dene their own tables

11
  three dierent kinds of SAP tables
transparent: mapped 1:1 to RDBMS tables

pool: mapped n:1 to RDBMS tables

motivation: in the 80s, some RDBMS products limited the
total number of tables
cluster: mapped n:1 to RDBMS tables so that related tuples of
several cluster tables are stored in one row of the RDBMS
table
motivation: sometimes good during transaction processing
Trend: make all tables transparent

12

Examples
1. All comments and descriptions need to be stored in separate
tables in order to keep information in dierent languages:
part(id, . . . )
part(id, . . . , comment) ) comment(partid, language, comment)

N.B.: Actually, SAP tables come with names such as KONV, STXL,
VBAP, VBEP, etc. Keys span several attributes (including business
unit, etc.)

13
 2. Generic way of dealing with pricing terms (customizable):
lineitem(id, . . . , pricing term id )
lineitem(id, . . . , tax, discount) ) pricing terms(id, condition, amount)
 tax and discount are stored in two dierent tuples
(additional pricing terms stored in additional tuples)
 to allow quick access, pricing term tuples that belong to the
same lineitem are clustered together (i.e., pricing terms is a
cluster table)

14

Schema: Observations
 SAP databases tend to be very large (due to genericity)
 Schema is the heart of SAP, but still under constant revision
{ a couple of thousand new tables with every new major release
{ a great deal of reorganization work with every upgrade

15
 SAP's Transaction Concept
 SAP uses the term Logical Unit of Work (LUW) for transaction.
 Basically, an SAP LUW has the same ACID properties as SQL
and any (SQL) database system:
{ an SAP LUW can span several dialog steps
{ an SAP LUW is either executed completely or not at all (i.e.,
atomicity)
{ ...
{ nested transactions are also possible

16

Overview of Implementation
 SAP LUWs are NOT mapped 1:1 to database transactions
 SAP implemented its own locking
(centralized enqueue service )
 basically, SAP also implemented its own TP monitor
(message handler and queues in every application server)
 online transactions and batch (overnight) queries possible
 (for comparison: PeopleSoft uses third-party TP monitors such
as Tuxedo)

17
 Processing Dialog Steps
1. when a user logs in, a message handler
Application Server nds the application server with the
smallest load (load balancing)
2. this application server handles all of the
requests of that user session
Dispatcher
3. a user session consists of several
Queue transactions, and every transaction consists
of several dialog steps
Work ... Work 4. every application server has one dispatcher
Process Process process and several work processes
5. the dispatcher queues requests until a work
private
buffers
private
buffers
process is available
6. a work process carries out a dialog step;
Shared Memory Buffers
rolls in relevant data and interprets
DynPro and ABAP programs
7. so, every user session is handled by a single
application server, but every dialog step is
handled by dierent work processes
8. exception: transactions that involve large
objects have exclusive work processes to
19 18 avoid cost of rolling data in and out
 lock requests lock release

P1 P2 P3 P4

posting steps

D1 D2 D3

dialog steps

online phase posting phase

 log records for updates are generated as part of every dialog step
 log records are propagated to RDBMS in posting phase
 locks are requested during online phase and released after the
posting phase is complete (2-phase locking)
 like dialog steps, posting steps are handled by dierent work
processes (potentially in parallel)
20

Why doesn't SAP directly use the RDBMS?

 Typical RDBMSs do not allow transactions that cross process
boundaries.
(This is necessary in SAP because the dialog steps of an LUW
can be handled by dierent work processes)
 aborts are quite frequent (e.g., out of stock) and only carried out
before posting; as a result, no roll-back at the RDBMS, the
bottleneck, is required for aborts
 SAP carries out locking in the granularity of \business objects"
which are dened in the ABAP dictionary

21
 Overview
 SAP's application servers cache ABAP programs, constraints,
and operational data in order to reduce the load of the RDBMS.
 more than 90% cache hits are not unusual for SAP applications
 ABAP programs and constraints are always cached.
 Administrator decides which data to cache and in which way to
cache it:
{ data that is frequently updated should not be cached at all
{ huge data that is likely to ood the cache should not be
cached either
{ default settings for pre-dened tables help
22

Caching Examples
 REGION table is a classic candidate for caching
 LINEITEM table is a classic candidate for NOT caching

23
 Application servers can cache data three dierent granularities
(set by default or system administrator for every table):
1. complete caching of a table
 cache can be used for any query on that table
 need much cache space, high cost to propagate updates
2. tuple-wise caching
 cache only used for select single statements
 ne-grained: i.e., good cache utilization, low cost in the
presence of updates
3. generic caching (a compromise)
 cache all tuples of a table with the same value of a prex of
the primary key
 e.g., cache all tuples that belong to the same business unit
24

Cache consistency in congurations with

several application servers
 periodic propagation
 no guarantee for cache coherency
 however, in practice not a problem because only data that is
almost never updated is cached or data for which inconsistencies
don't matter that much

25
  again, SAP implements its own authorization model and does
not use the standard (SQL) model supported by the RDBMS
 users must log in with their user-id and password (identication)
 ne-grained and exible authorization concept
{ individual elds of tables
{ specic transactions and/or reports
{ views
 bundling of authorizations
{ authorization object (set of related authorizations)
{ authorization prole (set of authorization objects; roles)
{ group prole (set of authorization proles)
 users belong to groups and inherit group authorizations
26

Security
 activity logs
 encryption of all messages exchanged between RDBMS and
application servers
 Kerberos and SecuDE for secure clients at the presentation layer
 secure transport system for batch input and migration of
databases
 only the administrator has direct access to the database and le
system; everybody else must use the interfaces of the presentation
layer or other external services in order to work with the system

27
  SAP monitors the following parameters
{ queue lengths in dispatcher
{ cache hit ratio
{ database operations (scans, sorts, joins)
{ ABAP operations (sorts, etc.)
{ number of commits and rollbacks
{ CPU, disk, memory, network utilization
{ response time of dialog steps, work processes, database calls
 alerters inform administrators if problems occur
 performance statistics are stored in the database
 EarlyWatch service | (Big Brother is Watching You)

28

SAP Transaction Processing Benchmarks

 dened in 1993 (Release 1.1H)
 purpose of these benchmarks:
{ sizing of an SAP system
{ measure real SAP application operations
{ key for database systems certication
 seven dialog benchmarks (FI, MM, SD, PP, PS, WM, ?)
 two batch benchmarks (AA, HR)
 most popular benchmark is SD benchmark
 usually run by hardware vendors
(SAP benets from their competition)

29
 The SD Benchmark
(Sales and Distribution)

Set Up (common to all benchmarks)

 one presentation server simulates users of several business units
 to avoid lock contention, maximum 100 users per business unit
 central conguration: one machine for application server and
RDBMS
 distributed conguration: n machines for application servers,
another machine for RDBMS

30

SD Benchmark Script
1. create an order with ve line items
2. create a delivery for this order
3. display the customer order
4. change the delivery and post goods issue
5. list forty orders
6. create an invoice
15 dialog steps; 4 posting steps; 150 secs think time

31
 SUN Beaverton, CA, USA on March 11, 1997:
Number of benchmark users: 1,410 SD
Average dialog response time: 1.85 seconds
Throughput: 7,133 SAPS (428,000 dialog steps/h)
Equivalent to: 143,000 processed order line items per hour
Average DB request time: 0.397 secs (dialog), 0.397 secs (update)
CPU utilization: 94%
Operating System: Solaris 2.5.1
RDBMS: Informix Online Server 7.21 UC1
R/3 Release: 3.0E (with 3.0F kernel)
Total disk space: 350 GB
1 Central Server: 64-way UltraSparc II CPU's 250 MHz, 14 GB main
memory, 1 MB level 2 cache; 15 dialog/update instances, 1
message/enqueue instance
Certication Number: 1997007
32

SD Benchmark Results and Trends

600000

500000
Transactions per hour

400000

300000

200000

100000

0
1993 1994 1995 1996 1997 1998
Year

hardware vendors have solved many of SAP's performance

problems concerning transaction processing.

33
  R. Buck-Emden and J. Galimow: SAP R/3 System { A
Client/Server Technology. Addison-Wesley, 1996.
 J. Doppelhammer, T. Hoppler, A. Kemper, and D. Kossmann:
Database Performance in the Real World { TPC-D and SAP
R/3. ACM SIGMOD, 1997.
 B. Lober and U. Marquard: R/3 System Standard Application
Benchmarks.
http://www.sap-ag.de/products/techno/index.htm

 B. Lober: R/3 System Standard Application Benchmarks {

Published Results, Version 2.4.
http://www.sap-ag.de/products/techno/index.htm

 SAP AG: Online Documentation. 1996-1998.

 SAP AG: SAP R/3 Software Architektur. 1994.
34

Queries and OLAP Applications on SAP R/3

Alfons Kemper

Outline
 Query interfaces of SAP R/3
 Decision Support Queries on SAP R/3
 SAP Data Warehouse Products

35
  all business data is entered via OLTP-applications into SAP R/3
 SAP has its own predened schema
 SAP R/3 is used to ask queries, e.g.:
{ for nding particular information as part of OLTP processing:
what happened to a particular order issued by customer
'Smith' in March 1998
{ for generating business reports for decision support:
how did our new line of products sell in dierent regions over
the last six months
 SAP R/3 has many predened business reports that can be
invoked \by the click of the mouse"
 SAP R/3 also allows to \program" customized queries/reports
36

Query Interfaces of SAP R/3

SAP R/3
ABAP/4 interpreter
Native SQL Open SQL
6 application- 6
SAP-SQL
data ?
SQL queries
DB-data

data - database
dictionary interface
local buers
6
? ?
relational database system
(back-end server)

37
 Open SQL
 SQL-92 style syntax (some limitations!)
 built-in operators, such as SORT
 materialize temporary results
 caching of query results
 cursor caching to optimize repeated database access
 in recent versions SAP has signicantly improved the database
interface
 before Version 3.0 it was not possible to \push" a join to the
database server; it had to be implemented as nested SFW's
within the ABAP-processor (i.e., within the application server).

38

Open SQL versus Native SQL{cont'd

Native SQL
 bypass ABAP/4 processor (and caching)
 bypass SAP data dictionary
 therefore not possible to access non-transparent tables
 advantage: often better performance
{ exploit (non-standard) features of the RDBMS
{ avoid certain problems of the ABAP processor
 disadvantage: not portable and potentially unsafe
{ cannot use dierent RDBMSs
{ programmer needs expert knowledge
) SAP recommends use of Open SQL
39
{ all built-in business applications use Open SQL
 40
The TPC-D Database Schema
PART (P_) PARTSUPP (PS_) LINEITEM (L_) ORDER (O_)







SF*200k SF*800k SF*6000k SF*1500k
PARTKEY PARTKEY ORDERKEY ORDERKEY

hardware and the same conguration

benchmark directly on commercial RDBMS using the same
for comparison: we implemented the \original" TPC-D
memory, 54GB disks, commercial RDBMS as back-end
Sun Sparc Station with 260MHz processors, 256 MB main
facility
implemented 2 update functions using SAP R/3's batch input
wrote 17 TPC-D queries in a) Native SQL and b) Open SQL
load data into SAP's predened tables
(Experience Paper). SIGMOD Conference 1997: 123-134
Performance in the Real World - TPC-D and SAP R/3
Hoppler, Alfons Kemper, Donald Kossmann: Database
results were reported in: Jochen Doppelhammer, Thomas
use of the TPC-D Benchmark
integer integer integer integer
NAME SUPPKEY PARTKEY CUSTKEY
char 55 integer integer integer
MFGR AVAILQTY SUPPKEY ORDERSTATUS
char 25 integer integer char 1
BRAND SUPPLYCOST LINENUMBER TOTALPRICE
char 10 decimal integer decimal
TYPE COMMENT QUANTITY ORDERDATE
varchar 25 varchar 199 decimal date
SIZE EXTENDEDPRICE ORDERPRIORITY
integer CUSTOMER (C_) decimal char 15
CONTAINER SF*150k DISCOUNT CLERK
char 10 CUSTKEY decimal char 15
RETAILPRICE integer TAX SHIPPRIORITY
decimal NAME decimal integer
COMMENT char 25 RETURNFLAG COMMENT
varchar 23 ADDRESS char 1 varchar 79
varchar 40 LINESTATUS
SUPPLIER (S_) NATIONKEY char 1
SF*10k integer SHIPDATE
SUPPKEY PHONE date
integer char 15 COMMITDATE
NAME ACCTBAL date
char 25 decimal RECEIPTDATE
ADDRESS MKTSEGMENT date
varchar 40 char 10 SHIPINSTRUCT
NATIONKEY COMMENT char 25
integer varchar 117 SHIPMODE
PHONE char 10
char 15 NATION (N_) COMMENT
ACCTBAL 25 varchar 44
decimal NATIONKEY
COMMENT integer REGION (R_)
varchar 101 NAME 5
char 25 REGIONKEY
REGIONKEY integer
integer NAME
COMMENT char 55
varchar 152 COMMENT
char 152

41
  8 tables in the TPC-D schema versus 17 tables in SAP R/3
SAP Table Description Orig. TPC-D Tab.
T005 Country: general info NATION
T005T Country: Names NATION
T005U Regions REGION
MARA Parts: general info PART
MAKT Parts: description PART
A004 Parts: terms PART
KONP Terms: positions PART
LFA1 Supplier: general info SUPPLIER
EINA Part-Supplier: general info PARTSUPP
EINE Part-Supplier: terms PARTSUPP
AUSP properties PART, SUPP, P.-S.
KNA1 Customer: general info CUSTOMER
VBAK Order: general info ORDER
VBAP Lineitem: position LINEITEM
VBEP Lineitem: terms LINEITEM
KONV Pricing Terms LINEITEM
STXL Text of comments all
42

Size of Database in KB (SF = 0.2)

Original SAP DB
TPC-D DB (Version 2.2)
Data 271,139 2,813,216
Indexes 102,822 841,008

Why is SAP database so big?

 SAP keeps info which is not modelled in the TPC-D benchmark
{ given default values in our experiments
 (strong) partitioning of the database tables
{ space to store foreign keys
 in SAP, keys (e.g. Customer-No) are 16 byte strings.
 in original TPC-D benchmark, keys are 4 byte integers.
43
 TPC-D Power Test Loading Time for the Benchmark
Query RDBMS Native SQL Open SQL Data (SF = 0.2)
Update (TPC-D-DB) (SAP DB) (SAP DB)
Q1 6m 09s 58m 59s 56m 18s Loading Time
Q2 53s 3m 09s 34s REGION |
Q3 4m 03s 9m 02s 11m 51s NATION |
Q4 1m 45s 6m 18s 6m 38s
Q5 6m 39s 14m 42s 37m 27s SUPPLIER 18m
Q6 1m 20s 7m 28s 14m 06s PART 15h 56m
Q7 9m 03s 23m 05s 29m 24s
Q8 1m 54s 19m 04s 16m 37s PARTSUPP 30h 24m
Q9 8m 42s 31m 33s 1h 7m 14s CUSTOMER 7h 33m
Q10 5m 18s 33m 06s 57m 49s ORDER+LINEITEM 25d 19h 55m
Q11 5s 4m 37s 2m 23s
Q12 3m 15s 9m 48s 9m 36s
Q13 8s 19s 25s  in all, it took about 1 month!
Q14 6m 23s 10m 25s 21m 54s  Reasons:
Q15 3m 25s 13m 51s 28m 31s
Q16 13m 24s 3m 16s 3m 22s { SAP carries out extensive consistency
Q17 11s 1m 50s 2m 13s checks for every record individually
UF1 1m 40s 1h 46m 54s 1h 46m 54s
UF2 1m 48s 11m 35s 11m 35s { SAP R/3 inserts records into the
Total (quer.) 1h 12m 37s 4h 10m 32s 6h 06m 22s database one-at-a-time rather than
Total (all) 1h 16m 05s 6h 09m 01s 8h 04m 51s making use of the RDBMS bulk load
facilities
 more details (all queries in Open and
Native SQL):  it appears to be very dicult (impossible?)
to bypass SAP R/3 and insert
http://www.db.fmi.uni-passau.de/projects/SAP/ R/3-consistent data into the RDBMS
45 44 tables directly
  three (native SQL) to ve (Open SQL) times more expensive
queries
{ ten times larger database (due to partitioning, large keys,
default values)
 partitioning of the data leads to N  M -way joins versus M -way
joins in original TPC-D benchmark
{ Example: Query 1
{ TPC-D: single table scan
{ SAP: Lineitem , Lineitem , Lineitem
 updates are very expensive because of the consistency checks;
just as for initial loading

46

Why is Open SQL even more expensive?

 Limitations of Open SQL prohibit \pushing down" certain
expensive operations, e.g., complex aggregation
{ increased communication costs between database server and
application server
 poor query evaluation plans
{ partly due to RDBMS
{ partly due to SAP R/3
 cannot exploit non-standard features of the RDBMS
{ for example proprietary string manipulation routines
 SAP generates parameterized queries rather than passing
constants of selection predicates to the RDBMS optimizer
{ may result in poor optimization choices by the optimizer
47
{ e.g., choosing an index scan instead of a table scan because
the selectivity of a predicate is not known due to param.
 Conclusions
 Complex queries for decision support are too expensive in SAP
R/3
 The highly partitioned schema is optimized for OLTP
applications, not for OLAP applications
 (Extensive) query processing penalizes the operational OLTP
system
) Data Warehouse is the way to go

48

SAP EIS (Executive Information System)

 SAP's rst attempt in data warehousing
 part of the regular R/3 system
 uses the same system (database server and application server) as
the OLTP applications
 therefore, incresed load on the OLTP system
 based on a rudimentary star schema
{ a fact table
{ several dimension tables
 however, SAP EIS has limitations: e.g., only three attributes per
dimension table

49
  limitations in query processing
{ only 8 of the 17 TPC-D queries could be expressed in SAP
EIS
 the performance of these eight queries improved by a factor of
two as compared to the Open SQL reports
 high loading time: about 50 hrs to load the TPC-D database
with SF=0.2 from R/3 into EIS
{ even though, EIS was part of the same R/3 system
 merely extracting the TPC-D database from SAP R/3 into
ASCII les takes 6 hours

50

The Business Information Warehouse (BIW)

 SAP's new data warehouse product
 rst beta tests; general availability starting in July
 ROLAP architecture (Relational OLAP)
 RDBMS independent, i.e., several RDBMS products are
supported as back-end server (developed on Oracle and MS SQL
Server)
 bundled with the SAP R/3 system, Version 4.0
 but can also be acquired separately as a stand-alone system
 rst SAP stand-alone component (outside of R/3)
 part of SAP's Business Framework
; componetization of R/3
 BIW can handle data extraction from various other systems, not
51
only from SAP R/3 and R/2
 The Business Information Warehouse's Role
Business Business Business Business
Explorer Explorer Explorer Explorer

Web

Business Information Warehouse Admministrator

Server Workbench

R/3 OLTP R/2 OLTP Other OLTP External Info.

(diff. versions) Applications Applications Sources

52

The BIW's Architecture

Business Explorer

Report Catalog Browser

Reporting and Analysis Reporting and

for Microsoft Excel Analysis

BAPI
BIW Server
OLAP Processor

Meta Data Meta Data Manager Data Manager InfoCubes

Repository

Staging Engine Data Store

BAPI

OLTP Production Data Non R/3 Production

Reporting Extractor Data Extractor

R/3 OLTP Applications Non R/3 OLTP Applications

53
  open Business Application Programming Interfaces (BAPIs) for
{ data loading: to accommodate the extraction of data from
non-SAP systems
{ OLAP processing: e.g., for third party visualization tools
 pre-congured meta data repository: InfoCubes catalog, report
catalog, information source catalog
 many pre-dened InfoCubes for common business applications:
e.g., market segment analyses, protability analyses, stock
inventory analyses, corporate indicator systems
 predened extraction routines to obtain this data from R/3
OLTP systems (with incremental updates!)

54

The Staging of Information

Business Information Warehouse

InfoCubes

Update Rules

Communication Structure

Mapping and Transformation Rules

Transfer Structure Transfer Structure

ALE (Application Link Enabling-Interface) ALE

Transfer Structure Transfer Structure

55
Extract Source Structure Extract Source Structure
 Administrator’s Workbench

Meta Data Scheduler Data Load Data Access

Maintenance Monitor Monitor

Meta Data Staging

Manager Engine
Data Manager

Statistics

 GUI-based
 sophisticated scheduler

56

InfoCubes: Star Schema

Agent

Product Customer
Sales

Outlet Time
 one normalized fact table
 several dimension tables
{ possibly non-normalized to model hierarchical characteristics
{ e.g., customers' city! county! country! region
57
  (Handy) Sales by Brand/Year/Country
Brand

00
195
0
all 450
2300
0
300
Nokia 600
0
350 Country
Bosch 500 0
850
0
Motorola 400 270 all
0
240 CH
Siemens 800 1200 1400 3400 34 0 0 Aus
D
Year
1995 1996 1997 all

58

Relational Representation of an \InfoCube"

Customer Master Data

Customer-No Name Name Location Industry-Key

InfoCube: Profitability by Customer/Product

C Customer-No Chain Office Head Office

C P T quantity sold revenue discount sales overhead stock value

T Period Fiscal Year

P Product-No Product-Group Brand Category

Product-No Indicator ...

Product Master Data

Product-No Language Product-Description

59
  slice and dice
 drill down/roll up

Region
er
stom
Cu
Region

Product Group

er
stom
Cu
Product Group

Region
er
st om
60 Cu
Product Group

Business Analysis Library

 aggregate functions: sum, min, max, average, . . .
 comparison functions: dierence, ratio, percent,, share,
correlation, . . .
 sequence functions: top/last n, cumulative sum, tertiles,
quartiles, ABC analysis, . . .
 exception conditions: absolute values, top/last n, top/last n%,
trends, . . .
{ used to highlight exceptions
 nancial functions: currency conversion, business period
conversion, . . .
{ very important for international enterprises

61
 Performance
 no benchmark results yet
 dedicated server
{ separate from OLTP server
{ can be congured as a three-tier client server conguration
(like SAP R/3)
 InfoCube summary levels: precomputed aggregations
 persistent report cache
 batch reports

62

Literature
 Jochen Doppelhammer, Thomas Hoppler, Alfons Kemper,
Donald Kossmann: Database Performance in the Real World -
TPC-D and SAP R/3 (Experience Paper). SIGMOD Conference
1997: 123-134
http://www.db.fmi.uni-passau.de/projects/SAP/

 SAP-AG: SAP Business Information Warehouse { Information

Center. http://www.sap-ag.de/products/biw/index.htm
 Rudiger Kretschmer and Wolfgang Weiss, SAP's R/3
Applications with ABAP/4. Sybex, 1996, Alameda, CA, USA.
 Transaction Processing Performance Council TPC, TPC
Benchmark D (Decision Support), Standard Specication 1.0,
Transaction Processing Performance Council (TPC), 1995,
http://www.tpc.org/

63