Professional Documents
Culture Documents
5A CCNIIDLN1IAL
1
InMemory Comput|ng (* denotes IN1LkNAL CNL quest|ons)
Jat |s meant by |nmemory comput|ng?
lnmemorv compuLlna leveraaes aroundbreaklna Lechnoloav lnnovaLlon Lo esLabllsh a
conLlnuous realLlme llnk beLween lnslahL foreslahL and acLlon Lo dellver slanlflcanLlv
acceleraLed buslness performance and reduced 1CC lnmemorv compuLlna comblnes boLh
LransacLlons and analvLlcs Lo dramaLlcallv lmprove exlsLlna buslness processes and enables a
whole new breed of lnnovaLlve new soluLlons lnmemorv compuLlna ls made posslble Lhrouah
recenL Lechnoloav advances ln maln memorv mulLlcore processlna and daLa manaaemenL
whlch comblne Lo dellver dramaLlc lncreases ln performance
Jat |s 5A's perspect|ve on |nmemory comput|ng?
Cnlv SA's approach unlquelv powers Lhe comblnaLlon of LransacLlonal and analvLlcal
appllcaLlons lL's more Lhan a new lnmemorv daLabase and fasLer analvLlcs SA ls Lhe onlv
companv LhaL can offer lnLearaLed sulLe of analvLlcal and LransacLlonal appllcaLlons LhaL
leveraae Lhe power of lnmemorv CompuLlna 8v aolna bevond [usL analvLlcs and [usL ln
memorv daLabases SA enables cusLomers Lo esLabllsh a conLlnuous realLlme llnk beLween
lnslahL foreslahL and acLlon ln pracLlcal Lerms Lhls equaLes Lo beLLer plannlna slmulaLlon and
forecasLlna beLLer sLoraae search and adhoc analvsls of Lodav's lnformaLlon comblnlna
sLrucLured daLa wlLh unsLrucLured daLa from Lhe web and enables mlxed workloads analvLlcs
performance manaaemenL and operaLlons ln a slnale svsLem
3 Jat are te benef|ts to customers of |nmemory comput|ng?
lnmemorv compuLlna empowers oraanlzaLlons Lo run fasLer work smarLer and perform
beLLer kun Iaster bv acceleraLlna buslness performance whlle reduclna 1CC Jork 5marter bv
enabllna emplovees and buslness neLwork users Lo make beLLer declslons fasLer and erform
8etter bv dramaLlcallv lmprove exlsLlna plannlna forecasLlna flnanclal close and prlclna
opLlmlzaLlon processes and bv enabllna whole new breed of appllcaLlons LhaL comblne
LransacLlons and analvLlcs
4 1|s sounds ||ke a |ot of vendor ype J|c rea| wor|d 5A customers are a|ready
us|ng t|s tecno|ogy and ow do tey benef|t?
SA lnLroduced 8uslness Warehouse AcceleraLor (8WA) ln 2006 LhaL showcased Lhe
performance beneflLs of columnar sLoraae over LradlLlonal rowbased svsLems lL has over 1000
cusLomers LlveCache was released ln 2002 and has over 2000 cusLomers CurrenLlv avallable
soluLlons LhaL leveraae lnmemorv compuLlna lnclude
O SA 8uslnessCb[ecLs Lxplorer acceleraLed verslon
O SA neLWeaver 8W AcceleraLor (formerlv called 8WA)
O SA C8M CusLomer SeamenLaLlon
O SA 8uslness 8vueslan analvLlcs
O SA Advanced lanner and CpLlmlzer wlLh llveCache
O SA neLWeaver LnLerprlse Search
IAC (Interna| d|str|but|on on|y)
5A CCNIIDLN1IAL
2
Jat |s te order of magn|tude of te |mprovements |n speed and vo|ume poss|b|e v|a
|nmemory comput|ng?
WhllsL lL has alwavs been deslrable Lo deal wlLh lncreaslna lnformaLlon volumes lL was noL
vlable ln Lerms of performance or feaslble from a cosL perspecLlve 1hls ls noL merelv
lncremenLal buL Lrulv aroundbreaklna lnnovaLlon as Lhe huae order of maanlLude
lmprovemenL ln daLa processlna speed (10000x lmprovemenL ln speed of access from dlsc Lo
memorv comblned wlLh masslvelv parallel mulLlcore processlna archlLecLures) and volume (1
1erabvLe blade) alona wlLh reducLlons ln cosL make lL vlable for Lhe flrsL Llme and feaslble for
mass adopLlon 1hls ls especlallv slanlflcanL as Lhe raLe of lncrease ln processor speed (Cu's)
has subsLanLlallv dlmlnlshed ln recenL vears so lmprovemenLs ln overall response Llmes need Lo
be found elsewhere
6 Can |nmemory systems support te needs of our |argest customers?
es absoluLelv! Plahlv efflclenL compresslon ls a kev aspecL of lnmemorv compuLlna and we
have alreadv seen a compresslon facLor of 23x aL cusLomers Lxamples of 8W deplovmenLs
ranae from 30 60 18 ln CC around 3033 18 ln 8everaaes and harma and 26 18 ln Pl 1ech
lurLhermore lnmemorv compuLlna scales llnearlv and fullv leveraaes parallellsm so much
laraer deplovmenLs are posslble lor example based on 4 (8WA svsLems) x 2(mlrrored) x
28(8lades) x 48(C8/8lade) 10732 18
ow muc does |nmemory comput|ng cost and wat spec|a| ardware |s requ|red?
1he appllcaLlons LhaL are acceleraLed bv lnmemorv compuLlna are prlced dlfferenLlv based on
Lhe funcLlonallLv Lhev offer CurrenL producLs are LhaL leveraae lnmemorv compuLlna such as
8WA can be dellvered as an appllance We expecL oLher soluLlons Lo be made avallable on
demand ln Lhe fuLure where lnmemorv forms an lnLearal parL of Lhe hosLed landscape and
prlclna Lhus removlna Lhe need for cusLomers Lo acqulre hardware Lhemselves
5A CCNIIDLN1IAL
3
Jat rea|wor|d examp|es of |nmemory transact|ona| app||cat|ons does 5A support?
We expecL Lo lnmemorv enable a number of kev LransacLlonal svsLems ln Lhe near fuLure
examples lnclude SA 8uslness8vueslan as well as our plannlna budaeLlna and forecasLlna
soluLlons lf we look aL our currenL producLs AC and demand plannlna lnmemorv powered
soluLlons LhaL are lnLearaLed Lhrouah wrlLebacks lnLo our LransacLlonal svsLems AddlLlonallv
SA 8uslnessCb[ecLs lannlna and ConsolldaLlon verslon for SA neLWeaver can leveraae Lhe
querv/read enhancemenLs ln 8WA 720
Does t|s mean I w||| need to m|grate my ex|st|ng systems to take advantage of |n
memory comput|ng?
CusLomers do noL need Lo chanae roadmaps Lo beneflL from SA lnmemorv lnnovaLlons
LxlsLlna cusLomers have opLlons Lo enhance exlsLlna lnvesLmenLs Lo leveraae lnmemorv
capablllLles ln a nondlsrupLlve manner new soluLlons wlll lncreaslnalv leveraae lnmemorv
capablllLles lor example cusLomers LhaL have lnvesLed ln 8l 8W were able Lo leveraae 8lA and
8WA ln a nondlsrupLlve manner LhaL leveraaes exlsLlna lnvesLmenLs veL provlded dramaLlc
performance lmprovemenLs
Jat w||| 5A board members be say|ng |n te|r 5app|re keynotes around |nmemory
comput|ng?
Passo laLLner wlll focus on lnmemorv compuLlna ln hls kevnoLe emphaslzlna Lhe slanlflcance
of Lhls paradlam shlfL ln Lechnoloav Pe wlll dlscuss fronL Lhe polnLs of vlew of user deslrablllLv
buslness value and vlablllLv and Lechnlcal feaslblllLv uemo wlll showcase how Lhe plannlna
process can be slanlflcanLlv lmproved when enabled bv lnmemorv compuLlna
J|c |ndustry |eaders and partners |s 5A co|nnovat|ng w|t around |nmemory
comput|ng?
We belleve sLronalv ln a heLeroaenous ecosvsLem and parLnershlp sLraLeav We are lnnovaLlna
wlLh a number of sLraLealc parLners ln Lhls area Some examples lnclude noL llmlLed Lo
rocessors (lnLel) Servers (Clsco uell l8M) SLoraae (LMC) neLwork (Clsco 8rocade)
vlrLuallzaLlon (vMWare MlcrosofL)
3Jat |s te |mpact of te 5ybase acqu|s|t|on on our |nmemory efforts?
1he acqulslLlon of Svbase wlll alve our cusLomers lncredlble value due Lo Lhe svnerales beLween
Lhe Lwo companles 1he LransacLlon wlll help companles run beLLer Lhrouah realLlme declslon
maklna made posslble bv Lhe comblnaLlon of SA buslness and analvLlcs sofLware and ln
memorv Lechnoloav and Svbase's moblle lnfrasLrucLure and daLa manaaemenL soluLlons 1he
move has a sLraLealc beneflL for SA and Svbase SA wlll be able Lo acceleraLe Lhe reach of
SA soluLlons across moblle plaLforms and reallze lLs lnmemorv compuLlna vlslon lL wlll drlve
hlaher user adopLlon of SA sofLware and unlock slanlflcanL buslness value ouL of exlsLlna SA
lnvesLmenLs lor Svbase breakLhrouah lnmemorv Lechnoloav from SA wlll provlde a dramaLlc
performance lmprovemenL Lo analvLlc processlna capablllLles Svbase's daLabase producLs wlll
be enhanced bv SA lnmemorv Lechnoloav 1he move wlll furLher spur producL lnnovaLlon and
svnerales lncludlna offerlna a powerful comblnaLlon of Lhe Svbase lC producLs wlLh SA's
columnbased/lnmemorv asseLs for fuLure SA appllcaLlons on premlse and on demand
IAC (Interna| d|str|but|on on|y)
5A CCNIIDLN1IAL
4
4Are tere any spec|f|c bus|ness processes or |ndustr|es |n w|c t|s tecno|ogy offers
s|gn|f|cant|y greater va|ue?
1hls ls a core Lechnoloav and lLs beneflL and usefulness ls appllcable across all appllcaLlon
scenarlos and lndusLrles Some example lndusLrv scenarlos lnclude
O edge fund trad|ng ana|ys|s |n I|nanc|a| 5erv|ces More real Llme plcLure for hedae
fund accounL manaaers Lo manaae exposure Lo currencles equlLles derlvaLlves and
oLher lnsLrumenLs for Lhelr hedae fund cllenLs upslde ls lmproved Lradlna success
for Lhelr hedae fund cllenLs and beLLer cusLomer lovalLv A lorLune 300 cusLomer ls
currenLlv lmplemenLlna Lhls scenarlo wlLh SA 8uslnessCb[ecLs Lxplorer
O kea| t|me p|pe||ne p|ann|ng |n C|| Gas and M|n|ng WlLh endconsumer lnslahLs
and raw maLerlals lndusLrles such as mlnlna or oll and aas Lracklna Lhelr plpellne
aaalnsL markeL prlces and commodlLv Lradlna real Llme
O Dt|||t|es DemandDr|ven roduct|on |ann|ng lnLernal plannlna daLa can be
comblned wlLh realLlme consumpLlon daLa from SmarL MeLers LhaL ls lnLearaLed
wlLh exLernal daLa such as weaLher daLa and evenLs so LhaL plans can be slmulaLed
wlLh dlfferenL opLlons
Jat support does 5A prov|de to partners and t|rd party vendors to |everage |n
memory comput|ng?
arLners and cusLomers can leveraae exlsLlna Als on 8WA Lo bulld or cusLomlze soluLlons
Lodav ln Lhe meanLlme we are movlna Lowards more open sLandards such as SCL Mux SA
wlll provlde aaLewav Lo expose servlces as a plaLform LhaL leveraae lnmemorv compuLlna
capablllLles
6Most ex|st|ng d|skbased systems a|ready do a |arge proport|on of te|r work |n
memory troug cac|ng ow and wy |s 5A's |nmemory comput|ng d|fferent and
better?
1hese are noL Lrue lnmemorv svsLems raLher LradlLlonal daLabase svsLems LhaL emplov
cachlna SA's lnmemorv Lechnoloav ls deslaned around an lnnovaLlve and hlahlv efflclenL
columnar sLoraae" model whlch offers fundamenLallv superlor performance Lo LradlLlonal
svsLems even when slanlflcanL cachlna ls performed
ow does 5A's approac to |nmemory comput|ng d|ffer from tat ava||ab|e from
oter vendors suc as Crac|e I8M and M|crosoft and wy |s |t super|or?
WhllsL oLher vendors have made foravs lnLo lnmemorv Lo speed analvLlcs SA ls Lhe onlv
vendor LhaL brlnas lnmemorv compuLlna Lo Lhe LransacLlonal world as well ln addlLlon all
uslna a slnale daLabase
Jat |mpact does |nmemory ave on 5A's partner re|at|ons|ps w|t oter |n
memory vendors suc as I8M 1eradata etc
We are commlLLed ln supporLlna all daLabase and Lechnoloav parLners We are commlLLed Lo
worklna LoaeLher wlLh l8M and 1eradaLa so LhaL our 8W soluLlons can lnLeroperaLe wlLh 3
rd
parLv daLa sLores (lnmemorv or oLherwlse)
IAC (Interna| d|str|but|on on|y)
5A CCNIIDLN1IAL
3
Jat |s te tecn|ca| descr|pt|on of an |nmemory database? ow does te concept of
an |nmemory database d|ffer from a c|ass|ca| database? Are you [ust us|ng kAM
|nstead of d|sk? Jy can I not [ust use a c|ass|ca| kD8M5 |n memory? Jou|d tat |ead
to te same resu|ts?
1he cenLral ldea of an lnmemorv compuLlna envlronmenL ls Lo reduce Lhe laLencv Llme for
brlnalna daLa Lo a processor uaLa needs Lo reslde wlLh Lhe Cu realsLers when an operaLlon ls
execuLed ln order Lo aeL Lo Lhese realsLers daLa needs Lo Lravel from lLs orlaln whlch could be
dlsk or 8AM When vou compare Lhe Cu Lo Lhls venue readlna daLa from 8AM ls llke walklna
from lrankfurL Lo Walldorf Lo slp some waLer readlna daLa from dlsk ls llke Lravellna Lo Mars
Lo reflll vour alass
Classlcal daLabases have Lrled varlous approaches Lo brldae Lhese daLa Lravel Llmes uslna
Lechnoloales llke cachlna preaaareaaLlons of Lables Lemporarv Lables ln memorv amonasL
oLhers Powever Lhe maln Lechnoloav has been wrlLLen 20 vears back wlLh Lhe dlsk ln mlnd as
Lhe maln sLore for daLa ConcepLs llke paaes exLenLs or Lemporarv Lable spaces are sofLware
arLlfacLs of Lhls area uaLabases are Lhe foundaLlon for Lhe currenL 3Ller archlLecLure of
buslness appllcaLlons where daLa resldes ln Lhe daLa Ller and appllcaLlon loalc provldes Lhe
appllcaLlons exposed ln Lhe presenLaLlon Ller on Lhe cllenL machlne 1hls archlLecLure ls known
as cllenL/server archlLecLure
ln memorv compuLlna ls based on Lhe ma[or Lechnoloav shlfLs happenlna Lodav modern
servers have manv Cus wlLh manv cores and Lhev are capable of addresslna huae amounLs of
affordable memorv lL ls posslble Lodav Lo run Lhe ma[orlLv of enLerprlses ln one slnale server
on 64 cores based on 218 of memorv
SA's lnmemorv Lechnoloav ls a new aeneraLlon of sofLware LhaL leveraaes Lhls new hardware
Lechnoloav aL full exLenL uaLa ls oraanlzed ln wavs LhaL make processlna ln memorv verv
efflclenL 1he columnar handllna of daLa enables slanlflcanL compresslon whlch leads Lo
efflclenL communlcaLlon beLween 8AM and Cus Avoldlna cache mlsses and uslna SlMu (slnale
lnsLrucLlon mulLlple daLa) capablllLles ln lnLel Cus furLher enhances performance
llnallv Lhe avallablllLv of full feaLured Cu cores enables complex operaLlons and calculaLlons
on daLa 1hese capablllLles provlde Lhe basls Lo reduce Lhe complexlLv of sofLware slanlflcanLlv
proarammlna concepLs for deallna wlLh daLa laLencv such as aaareaaLlons lnLernal Lables and
oLher arLlfacLs from Lhe daLabase area are no lonaer needed AppllcaLlons can reach ouL Lo Lhe
full scope of deLalled daLa ln memorv Lo aeL Lhe lnformaLlon Lhev need aL runLlme AppllcaLlons
can even deleaaLe daLa lnLense operaLlons enLlrelv Lo Lhe lnmemorv Lechnoloav and [usL walL
for a slan LhaL Lhe operaLlon has been execuLed successfullv
5A CCNIIDLN1IAL
6
name one wlLh Lhe cusLomer counLrv one wlLh Lhe producL and one wlLh Lhe sales kev flaure
All values ln one column have Lhe same daLa Lvpe and characLerlsLlc so Lhere are onlv 230
counLrles ln Lhe world lnsLead of sLorlna Lhe counLrv ln each record accordlna Lo Lhe Lvpe
deflnlLlon Lhe counLrv column can use heavv compresslon on Lhe daLa and sLore Lhe column ln
an opLlmlzed fashlon ln addlLlon a querv Lvplcallv does noL requlre all aLLrlbuLes of a record buL
onlv Louches on speclflc columns Looklna for sales over realons and Llme onlv Louches on Lhree
columns A columnar daLa sLore wlll onlv Louch Lhese Lhree columns Lo reLurn Lhe resulLs A row
sLore would need Lo fllLer records and Lhen read each record wlLh all Lhe aLLrlbuLes before lL
can reduce Lhe resulL seL Lo Lhose aLLrlbuLes LhaL are acLuallv requlred
Columnar sLoraae allows for parallel processlna (Lrend of mulLlcore processors mulLl Cu
blades and mulLlblade racks) leveraalna advances ln processor Cu and blade Lechnoloales
Cverall Lhe columnar daLa sLore can slanlflcanLlv compress daLa and lL ensures LhaL onlv
relevanL daLa ls Louched durlna querv execuLlon ln Lhls wav columnar daLa handllna ls kev Lo
efflclenL lnmemorv compuLlna
Jy |s t|s a trend now ow as tat become feas|b|e? Is t|s a new concept? Is |t
ardware tat makes |t poss|b|e? kAM pr|ces??
1he concepL of columnar daLabases ls noL new uramaLlc lnnovaLlons ln Lhe hardware secLor
provlde us Lodav wlLh a compuLe plaLform LhaL llmlLed Lhe handllna of daLa ln memorv ln Lhe
pasL MulLl core svsLems wlll soon have 64 cores per Cu a number LhaL onlv hlah
performance clusLers ln research had avallable ln Lhe pasL and memorv ls aeLLlna cheaper and
denser
ln addlLlon Solld SLaLe ulsks (SSus) lnslde Lhese servers can perslsL Lhe lnformaLlon held ln
memorv so LhaL even ln Lhe evenL of a power or hardware fallure no lnformaLlon ls losL SSus
have recenLlv become avallable wlLh sufflclenL daLa LhrouahpuLs enabllna Lhem Lo server as
perslsLencv Lechnoloav for lnmemorv compuLlna
1odav vou can buv one server wlLh 4 Cu's wlLh 8 cores each runnlna on 118 of memorv lf vou
buv 10 blade servers vou have a masslve parallel plaLform wlLh 320 cores and 1018 of memorv
Lhls one blade enclosure could run 640 processes ln parallel and Lhe columnar sLore can hold
Lhe equlvalenL of 3018 daLa Lhere are noL manv companles ln Lhe world LhaL have more daLa
ln Lhelr buslness appllcaLlons Some of SA's laraer cusLomers use less Lhan 3018 Lodav for Lhelr
8uslness SulLe daLa 8LlC8L Lhe 20x compresslon LhaL we Lhlnk we can achleve wlLh ln
memorv compuLlna
ow do we ac|eve x compress|on?
Columnar SLoraae lends lLself Lo beLLer compresslon Lhan row based sLoraae bv comblnlna
analvLlcal and LransacLlonal capablllLles lnLo one svsLem whlch removes Lhe need Lo sLore
redundanL daLa Manv of Lhe compuLaLlonal alaorlLhms move Lo Lhe daLabase / comblned wlLh
Lhe speed of sLorlna daLa lnmemorv removes Lhe needs for caches Speed of access Lo ln
memorv sLoraae and parallel processlna ellmlnaLes some of Lhe need for aaareaaLe Lables (Lhls
ls ofLen done for performance reasons slmllar Lo Lhe cache araumenL above)
IAC (Interna| d|str|but|on on|y)
5A CCNIIDLN1IAL
7
3Jy wou|d a customer want to use |nmemory comput|ng rater tan trad|t|ona|
kD8M5? Is te |ger performance not on|y re|evant |n a very sma|| percentage of use
cases? Cr |s tere any advantage bes|de speed?
1he new Lechnoloav enables a compleLelv new breed of appllcaLlons LhaL no lonaer
dlfferenLlaLe beLween operaLlons and analvLlcs 1radlLlonallv daLa had Lo be repllcaLed from
operaLlonal appllcaLlons Lo analvLlc envlronmenLs flrsL ln order Lo avold masslve performance
lmpacL from analvLlc appllcaLlons on operaLlonal appllcaLlons WlLh Lhls new Lechnoloav Lhe
movemenL of daLa ls no lonaer necessarv CusLomers can dlrecLlv run analvLlcs ln real Llme on
operaLlonal daLa ln order Lo lncrease efflclencv
1hls ablllLv enables new lnformaLlon rlch appllcaLlons whlch have noL been posslble ln Lhe pasL
due Lo complexlLv and performance consLralnLs AppllcaLlons can run complex calculaLlons llke
forecasLs or slmulaLlons aL human speed meanlna less Lhan a second whlch provldes buslness
users wlLh more flexlblllLv Lo follow Lhelr Lraln of LhouahL
4It must be extreme|y expens|ve to put a |arge companys database |nto kAM Jat
are typ|ca| database s|zes wat |s te compress|on rate re|at|ve to wat tese
compan|es use today? Jat |s te J cost for a typ|ca| LL customer?
8ased on Lhe developmenL of memorv prlces over Llme Lodav vou can buv memorv needed Lo
hold daLa from 70 davs for Lhe same amounL of monev vou could buv memorv a couple vears
back Lo hold daLa from 3 mlnuLes 1here ls a wlde ranae of dlfferenL volumes of daLa belna ln
use bv cusLomers Lodav however onlv a hand full of cusLomers would exceed Lhe maxlmum
volume of memorv vou can puL lnLo one server Lodav (218) 1hls sLaLemenL ls based on an
averaae compresslon raLe of 110 so one server can currenLlv sLore neLLo daLa up Lo 1018
1here ls a dramaLlc decllne ln memorv prlces maklna servers wlLh 312C8 or 118 easllv
affordable ln addlLlon secondarv sLoraae Lechnoloales llke SSu provlde Lhe posslblllLv Lo aae
cold daLa Lo fasL SSus so LhaL such daLa no lonaer occuples maln memorv
Jat are te key d|fferent|ators from c|ass|ca| kD8M5? 5peed? Compress|on?
ke||ab|||ty? Lase of database management? ow do customers coose? Jat makes
Crac|es kD8M5 so dom|nant |n te market?
1he kev dlfferenLlaLor ls lnmemorv Lhlnklna" AppllcaLlons wlll operaLe ln a dlfferenL wav as
Lhev wlll no lonaer be bound Lo deallna wlLh Lhe laLencv when accesslna daLa 1hls paradlam
shlfL wlll have slanlflcanL lmpacL on how people lnLeracL wlLh svsLems LhaL respond aL human
speed Lo anv lnLeracLlon poLenLlallv leadlna Lo alLered buslness models
As an example a companv belna able Lo run Lhelr dunnlna processes aL anv polnL ln Llme ln a
few seconds can poLenLlallv alLer Lhelr lnLeracLlons wlLh cusLomers
A companv havlna full lnslahL lnLo lnvenLorv and belna able Lo ad[usL forecasLs based on up Lo
Lhe second polnL of sales daLa mav poLenLlallv ad[usL Lhelr buslness model
IAC (Interna| d|str|but|on on|y)
5A CCNIIDLN1IAL
8
ManaaemenL meeLlnas wlll mosL llkelv chanae as lnformaLlon ls avallable aL Lhe speed of
LhouahL and declslons do no lonaer need Lo be posLponed due Lo Lhe Llme lL Lakes Lo aeL Lhe
relevanL daLa
CusLomers should chose based on Lhe added value an lnmemorv compuLlna based soluLlon can
provlde over a slmllar soluLlon runnlna on a daLabase
6Jat about 5As own database MaxD8? Jat |s |ts |story? ow many compan|es
are us|ng |t? Is |t rea||y compet|t|ve? Jy as |t not captured a |ger market sare so
far? Can |t e|p 5A wen |t comes to |n memory? ow?
SA has never Lrled Lo poslLlon Maxu8 aaalnsL commerclallv avallable daLabases 1herefore
markeL share ls noL Lhe rlahL parameLer Lo measure Lhe success of Maxu8 SA has used Maxu8
for lLs ondemand offerlna SA 8vueslan ln order Lo have a conLrolled and manaaeable
envlronmenL ln house
ln Lhe conLexL of small and medlum buslness Maxu8 ls an lnLeresLlna alLernaLlve for SA
cusLomers Lo reduce Lhe llcense cosLs aLLached Lo a daLabase underlvlna SA soluLlons 1odav
we have approxlmaLelv 3300 cusLomers wlLh round abouL 7000 producLlve daLabase lnsLances
LoaeLher wlLh developmenL and LesL lnsLances Lhe lnsLalled base sums up Lo approxlmaLelv
14000 daLabase lnsLances
1he daLabase knowledae collecLed ln Lhe Maxu8 Leam ls kev Lo Lhe developmenL of Lhe
perslsLencv laver for our new lnmemorv compuLlna uaLa held ln memorv needs Lo be
perslsLed ln order Lo ensure durablllLv of daLa Cur daLabase experLs have profound knowledae
ln deslanlna loaalna and daLa perslsLencv mechanlsms for enLerprlse arade envlronmenLs
Jat market s|ze cou|d |nmemory comput|ng ac|eve? 1ota| database market |s
est|mated to be approx $bn (Gartner IDC) wat percentage of te market
wou|d move to |nmemory comput|ng? Is t|s a potent|a||y s|gn|f|cant revenue
opportun|ty?
1he lnmemorv compuLlna Lechnoloav ls noL lnLended Lo compeLe ln Lhe daLabase markeL lL ls
lnLended Lo supporL appllcaLlons LhaL no lonaer Lhlnk dlsk" buL LhaL Lhlnk memorv" We are
almlna aL exLendlna our markeL leadlna poslLlon ln Lhe area of enLerprlse buslness appllcaLlons
wlLh new soluLlons exLendlna our currenL core buslness CusLomers wlll be able Lo exLend Lhelr
currenL lnvesLmenLs lnLo enLerprlse sofLware wlLh lnmemorv compuLlna provldlna new added
value
Jat are oter p|ayer's act|v|t|es ere? ow about Crac|e? ow about startups? ow
about I/patents? A|| free ava||ab|e?
As Lhe Cracle CLC has expressed ln recenL lnLervlews Lhev seem Lo noL have Lhe lnmemorv
Lechnoloav under conLrol Cracle has aoL sLrona rooLs ln Lhe relaLlonal daLabase buslness and a
revenue sLream Lo proLecL so we expecL Lhem Lo conLlnue on Lunlna Lhelr daLabase as
wlLnessed wlLh Lhe Lxaserver We are sure Lhev know whaL Lhev are dolna ln Lhe daLabase
IAC (Interna| d|str|but|on on|y)
5A CCNIIDLN1IAL
buslness however we expecL Lhem Lo have dlfflculLles Lo LranslLlon Lo an lnmemorv Lhlnklna
whlch ls mandaLorv for cloud compuLlna
1here are oLher polnL soluLlons LhaL baslcallv Lrv Lo address Lhe challenaes seLup bv LradlLlonal
Lechnoloales ln Lhe analvLlcs space 1hese producLs are all addlLlons LhaL repllcaLe Lhe daLa lnLo
a hlahlv opLlmlzed sLore for Lhe sole purpose of analvsls
SA ls ln a unlque as we alm aL brldalna Lhe aap beLween operaLlons and analvLlcs enabllna
cusLomers Lo run appllcaLlons and analvLlcs on Lop of Lhe verv same daLa ln one envlronmenL aL
an ouLsLandlna speed
SA's lnmemorv compuLlna pushes Lhe boundarles of lnmemorv Lechnoloav and our research
and developmenL work resulLed ln a whole sequence of paLenL flllnas ln manv areas of daLa
manaaemenL we provlde around breaklna new approaches whlch have been Lrlaaered bv new
hardware aeneraLlons avallable Lodav
ou cou|d argue tat oter p|ayers ave more exper|ence |n database bus|ness tan
5A wy sou|d you w|n t|s market? Jat |s 5As compet|t|ve edge?
We do noL araue wheLher oLher companles have beLLer experlence ln Lhelr respecLlve domaln
of excellence however we do belleve LhaL daLabase knowledae ls noL Lhe mosL slanlflcanL
domaln ln order Lo lnnovaLe ln Lhe lnmemorv area AcLuallv Lhe leaacv of mllllons of llnes of
daLabase code wrlLLen slnce 180 mlahL prohlblL LradlLlonal daLabase vendors from provldlna
new wavs of Lhlnklna ln Lhls lnmemorv compuLlna area lL mlahL be our unlque poslLlon noL
havlna Lrled Lo own markeL share ln Lhe daLabase space whlle havlna Lhe Lechnoloav whlch
Lodav provldes us wlLh Lhe ablllLv Lo Lhlnk compleLelv ouLslde Lhe box
1he relevanL parL for our lnsLalled base ls Lhe appllcaLlon and lLs seL of capablllLles provlded Lo
Lhe users We belleve LhaL we can lnnovaLe ln Lhls area wlLhouL dlsrupLlna Lhe buslness of
cusLomers wlLh new capablllLles becomlna avallable Lhrouah lnmemorv compuLlna
5A CCNIIDLN1IAL
10
5A CCNIIDLN1IAL
11
34I ave eard you are us|ng t|s tecno|ogy |n te context of 8yDes|gn? Is t|s true?
ow exact|y?
Passo wlll demonsLraLe a runnlna 8vueslan svsLem uslna lnmemorv compuLlna 8lahL now we
are uslna Lhls Lechnoloav ln 8vueslan Lo acceleraLe all analvLlcs ln Lhe svsLem 8vueslan
cusLomers can access reporL run analvsls and perform plannlna processes aL human speed 1hls
capablllLv ls provlded bv lndexlna Lhe buslness ob[ecLs ln memorv and bv runnlna all querles
and analvsls aaalnsL Lhese ob[ecLs ln memorv
3I ave eard you are p|ann|ng to move 5A 8us|ness 5u|te on |nmemory tecno|ogy?
Is t|s true? ow exact|y?
1he research performed bv Lhe Passo laLLner lnsLlLuLe demonsLraLed LhaL Lhere ls huae
poLenLlal ln runnlna buslness sulLe appllcaLlons ln memorv Cne of Lhe aspecLs ls Lhe sparslLv of
daLa lnslde Lhe appllcaLlon Lables whlch leads Lo a dramaLlc daLa volume reducLlon when
loaded lnLo our columnar sLore ln addlLlon havlna all operaLlonal daLa ln memorv enables
flexlble calculaLlon models Lo drlve reporLlna and analvsls feaLures WlLh lnmemorv compuLlna
we can remove Lhe need Lo dlfferenLlaLe ln onllneLransacLlon processlna (CL1) and onllne
analvLlc processlna (CLA) svsLems We expecL Lhls capablllLv Lo be verv lnLeresLlna Lo our
lnsLalled base
36Jat appens wen an |nmemory system sutdown unexpected|y |s a|| data |ost?
|e |f a|| data |s e|d |nmemory ow can |t be saved wen before te system suts
down and subsequent|y restored once te system restarts?
As well as sLorlna all lnformaLlon lnmemorv loas are saved Lo perslsLenL sLoraae such as SSu
CM eLc ln Lhe evenL of an unexpecLed shuLdown Lhe lnmemorv envlronmenL can easllv and
qulcklv be reconsLrucLed uslna Lhese loas Slnce daLa ls hlahlv compressed and fullv leveraaes
parallellsm lncludlna mulLlcore svsLem resLores wlll be slanlflcanLlv fasLer Lhan ln Lhe pasL
3Jat sou|d we te|| prospects/customers wo ask about te p|ann|ng scenar|o asso
and V|sa| w||| sow |n te|r 5app|re keynote?
1hls ls a concepLual sales plannlna scenarlo LhaL leveraaes lnmemorv compuLlna Lo be flrsL
dellvered ln Lhe conLexL of SA 8uslness 8vueslan and LaraeLed aL an SML audlence WhaL Lhev
wlll show ls noL a producL LhaL ls currenLlv avallable for purchase and lL does noL replace SA
8uslnessCb[ecLs lannlna and ConsolldaLlon (8C) whlch remalns our sLraLealc plannlna
budaeLlna and forecasLlna soluLlon
3Does t|s cange 5A's p|ann|ng strategy? Jat p|ann|ng so|ut|on sou|d we be
recommend|ng to customers at t|s t|me?
1hls does noL chanae SA's plannlna sLraLeav We conLlnue Lo recommend LhaL
cusLomers/prospecLs flrsL conslder SA 8uslnessCb[ecLs lannlna and ConsolldaLlon whlch ls
our sLraLealc plannlna soluLlon SA neLWeaver 8W lnLearaLed lannlna (8Wl) mav be
consldered for cerLaln cases (ex cerLaln publlcsecLor budaeLlna requlremenLs Merchandlse
AssorLmenL lannlna 1rade romoLlon ManaaemenL and cases where Lhe requlremenLs can
onlv be meL bv 8Wl) See Lhe SA LM roadmap for more deLalls
IAC (Interna| d|str|but|on on|y)
5A CCNIIDLN1IAL
12
hLLp//smarLdlrecLorvsapcorp3080/AsseLs/asseLepx?lde2f18c33b2484bb324
6cd0a8db3c1
3Jen w||| 5A 8us|nessCb[ects |ann|ng and Conso||dat|on be |nmemory enab|ed?
Cur lona Lerm sLraLeav remalns Lhe same and ls deLalled ln our LM roadmap (see llnk above)
Cur sLaLed aoal ls Lo dellver an lnmemorv enabled LM sulLe ln Lhe 20122014 Llmeframe LhaL
supporLs mulLlple deplovmenL opLlons lncludlna onpremlse and ondemand lurLher as parL of
Lhls sLraLeav we alm Lo dellver a slnale plannlna soluLlon ln Lhe same Llmeframe wlLh a slnale
Lxcel cllenL
4J||| |ncrementa| |nvestments suc as |nmemory support be made |n 5A NetJeaver
8J Integrated |ann|ng (8JI)?
es Whlle 8Wl wlll beneflL from beLLer performance wlLh Lhe 720 release of Lhe SA
neLWeaver 8W AcceleraLor (8WA) and 8Wl cusLomers wlll be able Lo use loneer for
plannlna on 8Wl bealnnlna ln C4 2010 Lhese enhancemenLs were done prlmarllv Lo help
move 8Wl cusLomers off Lhe 8Lx Lools whlch are currenLlv ln malnLenance mode 8Wl wlll
noL replace SA 8uslnessCb[ecLs lannlna and ConsolldaLlon verslon for SA neLWeaver as lL
supporLs dlfferenL Lvpes of plannlna scenarlos and ls deslaned Lo be owned and malnLalned bv
buslness users versus l1
4J||| 5A 8us|nessCb[ects |ann|ng and Conso||dat|on vers|on for 5A NetJeaver
|everage te |nmemory enancements |n 5A NetJeaver 8J Acce|erator (8JA)
?
es SA 8uslnessCb[ecLs lannlna and ConsolldaLlon verslon for SA neLWeaver can leveraae
Lhe querv/read enhancemenLs ln 8WA 720
4Is 5A 8us|ness |ann|ng and 5|mu|at|on (5LM85 8J85) be|ng end of ||fe'd? J|||
|ncrementa| |nvestments be made |n 8J85?
no Whlle noL offlclallv end of llfe'd Lhere wlll be no lncremenLal lnvesLmenL ln 8W8S
43Jen w||| te re|ease of 5A 8us|nessCb[ects |ann|ng and Conso||dat|on vers|on
for 5A NetJeaver and vers|on for te M|crosoft p|atform be out of kampDp?
1he plan ls for boLh verslons Lo exlL 8ampup (and enLer Lhe unresLrlcLed shlpmenL phase) ln
Lhe mld!une 2010 Llmeframe (Lhls ls sub[ecL Lo chanae)
44* Jat |s te d|fference between NewD8 and |nmemory comput|ng?
lnmemorv compuLlna refers Lo opLlmlzed and enhanced buslness processes posslble wlLh
svsLems and appllcaLlons LhaL leveraae lnmemorv Lechnoloav under Lhe hood newu8 ls an
lnLernal SA codeword for Lhe supporLlna daLa manaaemenL Lechnoloav behlnd lnmemorv
compuLlna newu8 ls noL a producL lL wlll noL be announced lL ls noL a Cnevolce approved
Lerm nor a realsLered SA Lrademark and should noL be dlscussed exLernallv
4* Jat |s k|ver?
IAC (Interna| d|str|but|on on|y)
5A CCNIIDLN1IAL
13
8lver ls an lnLernal codename for an Cnuemand laLform deslaned Lo supporL llahLwelahL
processes lL ls boLh a deslanLlme and runLlme plaLform slmllar ln concepL Lo soluLlons llke
lorcecom or Cooale App Lnalne Powever unllke Lhese svsLems 8lver has been deslaned from
Lhe around up Lo provlde seamless lnLearaLlon wlLh oLher SA soluLlons
46Are we def|n|te|y announc|ng tat at some po|nt our bus|ness process
transact|ona|/operat|ona| app||cat|ons w||| be runn|ng on |nmemory tecno|ogy? If
so ow do we and|e twopase comm|t ro||back etc
SLandard proLocols (AClu compllanL varlous LransacLlon models eLc) wlll be supporLed and are
opLlmlzed for our lnserL onlv model
4L|sted |n te mater|a|s |s tat "Lvent Ins|gt |n Cp8| demo pod# Lvent Ins|gt |s not a
re|eased or announced product (not t||| Aurora re|ease |ater t|s year) It |s terefore
perm|tted to d|scuss te concept be|nd |t but not to name |t
1hls ls noL veL lnmemorv enabled however we have plans Lo do Lhls ln Lhe fuLure
48J 3 |s ||sted as work|ng on |nmemory My understand|ng |s tat tat |s not te
case 1e confus|on may be tat w|t 8J 3 you can do federated ana|yt|ca| |ndexes
tat go |nto 8JA w|c of course |s |n memory 8ut 8J 3 |s not te |n memory
component |tse|f (costs extra 8JA)
1hls sLaLemenL ls Lechnlcallv correcL lL ls archlLecLed ln a wav LhaL beLLer leveraaes lnmemorv
capablllLles lL ls opLlmlzed for 8WA buL ls noL LeLhered Lo lL
4Jere can I f|nd more |nformat|on?
ConsulL Lhe followlna porLal paae ln parLlcular Lhe lnmemorv facL book
hLLps//porLalwdfsapcorp/lr[/ao/km/docs/lnfocenLers/WS20Cfflce20of20Lhe20C1C/r
oarams20and20ro[ecLs/lnMemorv20uaLa20ManaaemenL/lndex_newhLml