Reference Models for the Standardization and Automation of Data Warehouse Architecture including SAP Solutions

CIplus
Band 3/2018
Re e ence Models o he
S anda diza ion and Au oma ion o
Da a Wa ehouse A chi ec u e including
SAP Solu ions
Regys Mene, Ha mu Wes enbe ge , H oje Husic


1
Re e enceModels o  heS anda diza ionand
Au oma iono Da aWa ehouseA chi ec u e
includingSAPSolu ions

Au ho s:RegysMene,Ha mu Wes enbe ge ,H ojeHusic
1 Abs ac 
A chi ecu al ap oaches a e conside ed o simpli y he gene a ion o e-usable building blocks in he
ield o da a wa ehousing. While SAP’s Laye Scalable A chi ecu e (LSA) o e s a e e ence model o
c ea ing da a wa ehousing in as uc u e based on SAP so wa e, ex en ed e e ence models a e
needed o guide he in eg a ion o SAP and non-SAP ools. The e o e, SAP’s LSA is compa ed o he
Da a Wa ehouse A chi ec u al Re e ence Model (DWARM), which aims o co e he classical da a
wa ehouse opologies.
2 In oduc ion
G owing in o ma ion needs o as changing business models equi e a s able bu scalable and lexible
da a wa ehouse (DWH) in as uc u e. Many companies s a ed hei DWH a chi ec u e de elopmen
ei he wi h a cen alized da a eposi o y and da a ma s buil as desc ibed by Bill Inmon (Inmon, 2005)
o wi h a dimensional da a model as desc ibed by Ralph Kimball (Kimball, 2008). Since hen, he
in as uc u es ha e e ol ed e olu iona ily and ha e become complex and hyb id based on a ious
a chi ec u al pa e ns. He e, e e ence models can help o ame he DWH s a egy.
Szwed, Komna a and Dymek published an on ology called DWARM desc ibing classical DWH
a chi ec u e s yles (Szwed, Komna a, & Dymek, 2015). They p opose a laye -based e e ence model
wi h da a con aine s like a empo a y s aging a ea, a cen al da a eposi o y and da a ma s and da a
con aine s like ex ac ion, loading and da a ma eeding (see Figu e 1). DWARM uses axonomies o
o ganize he co e componen s o he DWH in as uc u e e e ence model. A con aine class is
conside ed as an abs ac ion o componen s able o s o e da a. A p ocess class is seen as an
abs ac ion o he da a p ocessing asks. These wo pa en classes a e used wo build up wo
axonomies. Bo h axonomies a e ela ed because da a con aine s a e used as sou ces o a ge s o
he da a p ocessing uni s.


2
Szwed and his co-au ho s poin ou how hei e e ence model co e s gene ic classical opologies like
Kimball o Inmon. Fo example, he le model o Figu e 2 e lec s Inmon’s hub and spoke app oach in
a DWARM-based no a ion. Due o i s gene ic cha ac e DWARM can desc ibe hyb id a ian s, oo.
Limi a ions a ise wi h he assump ion o he o de o he laye s. Fo example, he Da a Vaul
a chi ec u e s yle subs i u es he ETL p ocess by ELT and he business ule ans o ma ion a e applied
la e in he da a manipula ion. Asi p oposes an ex ension o he DWARM app oach o co e Da a Vaul
(Asi , 2017). The Cen al Da a Reposi o y should be sepa a ed in sub-laye s Raw Da a Vaul and
Business Vaul as designed by Lins ed and Olschimke (Lins ed & Olschimke, 2016). The
Figu e 1: Laye s (g ey), da a con aine s ( ec angles) and p ocesses (ellipses) building
he co e elemen s o DWARM (see (Szwed, Komna a, & Dymek, 2015))


3
ans o ma ion p ocess has o be di ided in a syn ac ical s ep assigned o he s aging a ea and a
syn ac ical s ep (business ules) shi ed o he cen al da a eposi o y laye (see igh pa o Figu e 2).

Figu e 2: Inmon’s DWH Re e ence A chi ec u e Model (le ) and wi h Da a Vaul ex ension ( igh ) acco ding o he DWARM-concep (see
(Szwed, Komna a, & Dymek, 2015)



4
3 Compa isono SAP’sLSA oDWARM
SAP o e s wi h he Laye ed Scalable A chi e cu e (LSA) a DWH a chi ec u e e e ence model, oo.
Based on his e e ence model SAP’s Business Wa ehouse solu ion p omises a high deg ee o
s anda diza ion and au oma ion o he DWH/BI p ocesses, while no losing he ocus on he scalabili y
and lexibili y o he DWH/BI p ocesses, see (SAP AG, n.d.- A), (SAP AG, n.d. - B). In a big pic u e,
LSA is composed o wo main a eas: he En e p ise Da a Wa ehouse laye and A chi ec ed Da a Ma
laye (see Figu e 3.) Addi ionally, each o hese wo laye s con ains u he sub-laye s. In he Da a
Acquisi ion Laye , which is he en y poin o he da a in he SAP BW en i onmen , he sou ce da a is
s o ed in he so-called Pe sis en S aging A ea (PSA). Da a cleansing and con o ming ans o ma ions
a e pe o med on i , be o e da a a e loaded in he a ge objec s o he Da a P opaga ion laye ,
Co po a e Memo y o e en Ope a ional Da a S o es. The LSA’s Da a Acquisi ion laye oge he wi h
he Quali y and Ha moniza ion laye and hei da a loading p ocess would align o he DWARM’s
ex ac ion and loading p ocess. Fu he mo e, he LSA’s PSA p o ides he same unc ion as he
DWARM’s Tempo a y S aging A ea (TSA).
A e he da a loading p ocess is execu ed in SAP BW, da a is loaded om PSA o he Da a P opaga ion
laye . In ac , he main pu pose o he Da a P opaga ion laye is o in eg a e da a ha comes om
mul iple sou ces in o he so-called DWH. The e o e, he Da a P opaga ion laye is ac ually he DWH
laye o he LSA a chi ec u e and is compa able o DWARM’s Cen al Da a Reposi o y laye , as bo h
laye s se e o he consolida ion o he da a uploaded o he DWH. On he o he hand, be o e loading
da a o he Da a P opaga ion laye , da a could be also loaded in he Co po a e Memo y laye , which
may keep he his o y o da a ex ac ed o he DWH o a longe pe iod (Paleka , Pa el, & Shi alka ,
2015). A he i s glance a compa able elemen is no speci ied in DWARM. I can be a gued ha
DWARM’s Cen al Da a Reposi o y laye can co e LSA’s Co po a e Memo y laye . Howe e , a
coun e -a gumen would be ha DWARM’s Cen al Da a Reposi o y Con aine is seen as he single
sou ce o u h o he DWH while he LSA’s Co po a e Memo y laye is jus a ep oduc ion o he
s aging a ea’s (PSA/TSA) s o ing da a pe sis en ly o a longe pe iod o ime. Ne e heless, depending
Figu e 3: SAP BW Laye Scalable A chi ec u e (LSA and LSA++) (SAP Online Lib a y, 2017)


5
on he in e p e a ion o his a guemen , he e migh exis he possibili y o ex end he DWARM wi h an
addi ional Co po a e Memo y laye .
In o de o suppo he ope a ional epo ing needs o he di e en manage ial le els, da a om he
Da a Acquisi ion laye can be also loaded di ec ly o he Ope a ional Da a S o e laye . This is simila
o he Da a P opaga ion laye . The e o e, pa icula ans o ma ions can be applied on i and make i
eady o ul ill he daily decision-making p ocesses. He e, a di e ence om he Repo ing laye is ha
da a ha e a sho e ime span and se es a he o answe he eal- ime epo ing equi emen s o he
end use s. In addi ion o ha , ano he di e ence is ha he objec s on his laye do no ha e a s a
schema s uc u e as hey ha e in he Repo ing (Da a Ma ) laye , bu a la able s uc u e. This means
ha wi h big amoun s o da a he epo ing pe o mance can decline. Figu e 4 isualizes he men ioned
ex ensions o DWARM highligh ed on blue .
Figu e 4: Mapping o LSA on an ex ended DWARM e e ence a chi ec u e


6
Going back o he LSA, u he logical agg ega ions can be pe o med on he de ailed da a in he Da a
P opaga ion laye . These agg ega ed da a is loaded in he epo ing laye , whe e i can be u he
accessed by que ies. The e o e, da a on his laye is agg ega ed and s o ed pe sis en ly in di e en
SAP BW objec s (e.g. In oCubes e c.), which consis o mul idimensional (MD) s uc u es. Seen om
he DWARM laye s pe spec i e, he LSA’s Business T ans o ma ion and Repo ing laye s comp ise o
he same unc ionali y wi h he DWARM’s Da a Ma Feeding p ocess and Da a Ma con aine .
A las , he i ualiza ion laye suppo s he union/join o di e en epo ing laye ’s objec s in o a i ual
one. Thus, he only di e ence o he i ualiza ion laye wi h he epo ing one is ha da a is accessed
on he un ime i ually wi hou being s o ed pe sis en ly. When conside ing his laye om he DWARM
laye s pe spec i e, one can ealize ha i pe o ms he same unc ionali y o he DWARM’s da a
in eg a ion p ocesses and Fede a ed Da a Reposi o y con aine .
Finally, DWARM con ains he wo addi ional laye s Da a Deli e y and Cus ome Applica ions, which
a e no included in he LSA e e ence a chi ec u e’s igu e abo e, bu a e ac ually equi alen o he
equi ed SAP BW que ies (known as BEx Que ies), which a e hen needed o expose he da a o
di e en BI/BO applica ions. Ne e heless, when placing agains he SAP BW’s LSA and LSA++
a chi ec u es agains each o he s, besides he s uc u al di e ences ha hey ha e wi h each o he , in
o e all, he di e ence in con ex s be ween LSA and LSA++ a e no ha big. Howe e , i is wo h
men ioning ha he LSA++ is he ype o e e ence a chi ec u e ha aligns bes o he agile DWH
modeling. The e o e, wi h i s upg aded me hods o DWH modeling and i s enhanced objec s, i is now
easie o apidly build new da a lows and a he same ime, o e be e da a loading and epo ing
pe o mance.
As i can be seen in Figu e 3 SAP BW has added u he componen s and seem o ha e made he
LSA++ a mo e ad anced e sion o i s ini ial LSA e e ence a chi ec u e. Mo eo e , om he
implemen a ion pe spec i e, LSA++ uses enhanced SAP BW objec s ha ha e be e loading and
accessing pe o mance.
Pa icula ly, when jumping o he Open Ope a ional Da a S o e (ODS) laye , i pe o ms he same
unc ions as he LSA’s Da a Acquisi ion and ODS laye . As such, he OpenODSView objec , ypically
used in his laye , is a i ual objec ha does no s o e any da a pe sis en ly. Howe e , i may also be
used o gene a e pe sis en Da a Sou ces o s age da a physically in SAP BW on HANA (SAP AG,
n.d.- A). On he o he hand, i can ex ac da a i ually om he SAP HANA DB and b ing i in he SAP
BW on HANA en i onmen . F om he e, his i ually ex ac ed da a can be hen combined wi h
pe sis en In oP o ide s in he SAP BW using Composi eP o ide s, which can be la e hen used o
epo ing. Fu he mo e, i is also possible o build que ies o epo ing di ec ly on he OpenODSView
objec . This added unc ionali y in LSA++ e e ence a chi ec u e enables hen he eal- ime epo ing
o he da a di ec ly a he Open ODS laye wi hou any da a la ency. Thus, he ini ial unc ionali y o
LSA’s ODS laye is now ul illed by he LSA++’s Open ODS laye and i s OpenODSView objec . A las ,
his laye p o ides he necessa y capabili ies o sha e he SAP BW on HANA In oP o ide s as iews
in he HANA DB en i onmen and ice- e sa, HANA DB iews in o SAP BW on HANA (Haup , 2012).
In his way, use s ha e he oppo uni y o wo k wi h hei DWH models using he ha d-coded SQL
que ies in HANA DB o he p o ided SAP BW on HANA in e aces.


7
Following he upcoming da a low modelling, when compa ing u he LSA++ laye s o he ones o LSA,
no so many di e ences a e iden i ied among hem. Up o he A chi ec ed Da a Ma laye
(co esponding o he Repo ing laye in LSA) e e y hing is almos he same. On he o he hand, he
main di e ences he e eside in he enhanced SAP BW on HANA objec s used in he SAP BW on
HANA (LSA++). As such, he ADSO is conside ed as a uni e sal objec ha can be used in mul iple
laye s such as he Co po a e Memo y, Da a P opaga ion laye e c., bu wi h di e en se ings in i s
de ini ions o each laye . Fu he mo e, in he LSA++’s BW Vi ual Da a Ma laye , he
Composi eP o ide objec subs i u es he unc ionali ies o he p e iously used objec s (e.g.
Mul iP o ide e c.) o he Vi ualiza ion laye in he LSA. Mo eo e , an addi ional imp o emen o he
LSA++ he e is ha he Composi eP o ide is able o i ually in eg a e and epo on SAP BW
In oP o ide s om any laye o he LSA++, wi hou losing he epo ing pe o mance, as i would
no mally happen in he SAP BW on any DB (LSA) using Mul iP o ide objec s. A las , he LSA++`s
Flexible Da a Ma s/BW Wo kspace laye , signi ies he agile capabili ies o e ed by LSA++. Thus, da a
om he ac ual SAP BW models, can be u he exposed and ex ended using local o depa men al
da a o ul ill a depa men ’s agile and ad-hoc epo ing needs (Peleshuk, 2015).
A e ha ing a deep o e iew o he DWARM and LSA/LSA++ e e ence a chi ec u es i is o in e es
o e alua e he deg ee o s anda diza ion and au oma ion p o ided by SAP BW owa ds gene a ing
di e en DWH a chi ec u al models. Table 1 p o ides an o e all compa ison be ween DWARM and
LSA/LSA++ e e ence a chi ec u es.
Gene ally speaking, he LSA and LSA++ e e ence a chi ec u es ha e a high simila i y wi h he
DWARM one. Howe e , he LSA pa icula ly has wo mo e laye s (Co po a e Memo y and Ope a ional
Da a S o e) compa ed o he DWARM. On he o he hand, he LSA++’s Flexible Da a Ma and BW
Wo kspaces laye is no appended o he DWARM laye s, as i is oo speci ic in compa ison o he
o e all pu pose o DWARM i sel . Ne e heless, ega ding he abo e-sugges ed ex ensions in he
DWARM laye s, he au ho s o i also admi he ac ha hei e e ence a chi ec u e model is mainly
p o ided on he gene ic le el and has place o u he cus omiza ion wi h ega d o speci ic DWH
a chi ec u es (Szwed, Komna a, & Dymek, 2015). On he o he hand, an a gumen o he ex ension
o DWARM wi h ega d o he LSA and LSA++ laye s may ollow om he ac ha he ac ual pu pose
o LSA and LSA++ is o i a bes he capabili ies suppo ed by he SAP BW. The e o e, in compa ison
o DWARM, one could see he SAP BW’s LSA and LSA++ e e ence a chi ec u es in a le el down in
g anula i y.
None heless, he ollowing sugges ed laye s o be ex ended o DWARM a e no only ound in he LSA
and LSA++, bu may also be use ul in he cases o he abo e-men ioned DWH a chi ec u es. As such,
in hei book abou hei DWH app oach, Kimball & Case a (2004) iden i y he unc ionali y o he LSA’s
and LSA++’s Co po a e Memo y laye wi h he ones o long- e m a chi ing o he s aged da a. Indeed,
his is also he pu pose o he LSA and LSA++’s Co po a e Memo y laye . Thus, he au ho s sugges
ha s aged da a should be a chi ed in a long- e m eposi o y (in his case, in he Co po a e Memo y
laye ) up o he momen ha i is su e ha i will no be needed anymo e (Kimball & Case a, 2004, S.
8).
