Chapter 21: Parallel Databases

صفحه 1:
حجه() اطل-۳۰۵) :00 !0 

صفحه 2:
+ Obnper 00: Padi! Oukbwes مسلسی له ‎VO‏ Aterquery Pardietow Aoarcvery Pardew ‎(Herb‏ مووي وا وه( میا سیر لهج مج ‎Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 همه‎ 

صفحه 3:
محا لح + لاه اجه مت عقی مسا چم ان لو ۲ 4 ۱ © Repent decking orputers Peuture cvuliple processors ued this tread is ‏اه با سور‎ ۲ ‏ره موی و ی(‎ kaye © ‏دمل مصاصمصصس خأن وسوس ام جروا‎ ore oolevied ued sored Por hier ordlsir. ‏سل و اوه راما بو سوم سا سرا مج‎ BE borgesrde parcel chicbase sysiews ‏ای اس‎ Port © ‏سا بو‎ vokaves of dota ۶ ‏مومس مومس مسر‎ queries © ‏مس وكا تجوايعحصها كلها بوملخصم‎ سا0 لح 0 لا سواه 1 همه 0 

صفحه 4:
@ardebsw ۲ ‏ححصولت()‎ ۴ ‏عمط مه ام با مه ما‎ dike Por parcel VO. 0 امس ام روط ‎dota coc be ponticosd ond euch processor coe work‏ © ‎pw 0‏ © Qheres oF expressed fa high level loon (GGL, trocstated to rebatiocral ukyebra) © coches pordetzation easier. © OO PPered querer coo be roo to pordiel wits each oer. Orwurewy varied! toes core of ‏ای‎ © The, dotcboses coturdly lead thecweebes to pardiete. Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 ممه‎ 

صفحه 5:
VO Cardeisw مسوم رجا حاص وخا ‎Recher the eve required to retteve rehire:‏ 19 لسلس و من ما با ‎Bh‏ ۱ (Lortgootdl portiicoiay — tuples oF a rebica ore divided ‏حاصب ۳ اس جحاصك روم پمو‎ tuple resides vo oor cist. ‎of chokes =‏ نادصي ) ‎BL Porticaicn teckoiques‏ :ملس او ال وا مس ما یی ید | ‎Gerad the‏ سوم ار لحاس رم اه اه و و و سول ۶ ‎Raton fi roe D...0-‏ سا ان ۰ ‎© bet idewwte resul oF kesk Pucwioa fopphed i he portiiccicn, otdbute ‏وه سا‎ tuple. Gerd tuple to disk & ‎ ‎ ‎Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ee ©Sbervehnts, Cork ced Cnakershe 

صفحه 6:
+ 4/0 Padstow (Ova) ل سا هی نا تم ج؟ ۲ اه موم ‎Choose os otirbute os the‏ © ۳ [چی رت بت ریت ‎pontticcing vevior‏ © © © Let be he partion cinbute ude of «tuple. Puples suck that 1; S rgg 7 todek 7+ ۰ Duper wah ‏عد‎ > 1, 07 to deo O ond hires wth 7 > 1, ‏طاص‎ ‎tok rfl. فق خا صلب ‎(Ud), «tuple wks particcieny ottbute‏ 6[ ی پم و ‎wah‏ وق چات ‎ull ty bok 00, «1 Ape wats uke © wl 9p to desk, whe 3 tuple‏ ‎inke CO wilep i dekO.‏ Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ee ©Sbervehnts, Cork ced Cnakershe 

صفحه 7:
وا یگ در ون + ی اس ماه هی سا بممممو۵ :1 لت ۵2 ره ۶ تسه ه لت عطا تاه ی و ‎oll tuples suck thot the vdue oP‏ بدتدصورا. © ‎queries.‏ و — سر .06 > هع > 00 .© * سا0 لح 0 لا سواه 1 مه 0 

صفحه 8:
+ Oowpwisoa oP Portiociny Packages (Con.) © pot ‏لحري‎ Por sequecid scan oF ‏من ماس هی‎ rack query. © @lidsks hove ckovst oo equal cucpber of tuples) retrieval work is thus well ‏ما لول‎ disks. 11 Rone ‏فا بت سوه‎ proces © We vheterke pkey ore soutered unrees ul doko Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. oe ©Sbervehnts, Cork ced Cnakershe 

صفحه 9:
ee ©Sbervehnts, Cork ced Cnakershe + Cowpartera oP Particaay Techaquee(Ovn.) Areek portiiocicrs: © Good ‏اوه‎ access © @ssunnieg has Pucca is qood, cod porttivoicy otrbuies Porc a hey, tuples wail be equaly destibuted beter disks © Retard work hea wel boknerd betwera debe. ۲ Gord Por ‏اه وم من و حدم‎ © Caachi side doh, lean there avakble ‏و اه متسه و‎ ی له ماما موی بحاصل و اما سا من اه روم لیا 8 اه ‎wore‏ ۲ ‏له مه ,ماه ون‎ to ower rooge queries Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. 

صفحه 10:
+ Ovwparecn oP Partiociag Tevkdques (Ovc.) Rene porticiny: Provides deta chester by portiociony ctr vo. (Gord Por ‏سوه سوه‎ (Bord Por pot queries a portinciny inbute! ody ou dok ceeds ty be coveverd, (Por reek queries oo pontiivotoy atibute, var too Pew disks way ured to be ‏و‎ © Rewaisiog disks ore available Por vier queries. © Booed P result tuples ure Prow vor to a Pew blocks. © 12 ‏تيه‎ blocks ore to be Petched, they ore oil Petched Pro vor to Pew disks, ood poteatdl pardkelicw ‏ال و‎ access te uceted * Cxecople oP executiva shew. Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. eae ©Sbervehnts, Cork ced Cnakershe 

صفحه 11:
وت هب و ۱ ] rettiog too state cist. Bo bone rettives ore prePercbly portivced ucrvss ull he avatuble disks. BR a rekon cette of «7 dek boche ond here ore dike ‏ره‎ to the sume ‏او )مي اس سا امه ما ما‎ Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ence ©Sbervehnts, Cork ced Cnakershe 

صفحه 12:
+ ‏وسور‎ oP Clow BD he distrbutivg oF tuples to dishs way be skewed — thal is, svwe disks hove * Gowe udues oppeor to the portico otiributes oP acay tuples; oll the tuples wil the sawe vdlue Por the porticcice, otdbute eed up ic he sare porta. » Coa porwr wil ‏ولج روموت‎ © Comer skew. * Ol ree-ponticcicy, body choseo ‏ماو‎ verter ‏رو ما مه بو‎ ‏و با لد‎ portizes ood tow Peuy to ‏واه‎ ۱ bess they wih hosk-ponticatog Po good ‏واه و میا‎ Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 ممه‎ 

صفحه 13:
+ ALeediay Gkew ia Racne-Pariiocicry BP orede a bekneed portions vevior (seexrk ‏صطه سم‎ Borers 3 key of te rekon): © Gor the relics va he portiiccio citric. © Corstrunt the portiicg vevior by scosatay ihe retaica it sorted order us ‏تساو‎ * ORter every U/c8 of the retatiog hos bero read, the ude oP the portiicciog vttribute oP the ‏موم‎ tuple te added to the portiiza vector. © odeutes the anvber of portions to be coostrurted. © Ovpticdte cuites or kvboraves can rest P duplcdes ore preset ‏روم و‎ ‏.وج جا قد‎ مس ذا لحب وص جاه جب لیا سم شح صا 19 0 a0 ©Sbervehnts, Cork ced Cnakershe 

صفحه 14:
+ Ninny Glew ern beogace BE Ochre parte ‏اه و‎ remand Pro herman ‏امه راما وم‎ ۶ ‏بط موی و‎ idk euch rane of te Keira ‏و۱‎ ua be ‏سا باه را پم بجا لت‎ cout) tuples of te retain frequency 1-5 6-10 11-15 16-20 21-25 55 ka Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. enor ©Sbervehnts, Cork ced Cnakershe 

صفحه 15:
Warndiag Ghew Osiay Ortud Provessor Cortiodag Bl Chew terrane portioctag con be koaded elec uskry viral processor ‏سس‎ ‎۶ ‏بوم) وج اه واه من‎ UD ۰ 60 ‏ام سا سود‎ of ‏اس‎ ‎© poke vind processors i partion ether in rannbrobis Paskira or bosed pa esikated ovst oP provesiey park virked portinn ۲ 8 ‏حص‎ eu! © 9P oy word poniicg would have bees skewed, itis very tkely the shew i> spread over a cuober oF virtual ‏او‎ © Chewed virted portions yet spread uorvss uo ‏اه طلسم‎ processors, 57 work ype distiribuied evedy! Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. a0 ©Sbervehnts, Cork ced Cnakershe 

صفحه 16:
محاط دن ی + ای و ‎ta pordel wits‏ وت ساسح «أجتص 9 19 ماه و و و ار لح زوا مش 1 ۳ ‎oP irnsurines per seven.‏ وی ها نموه وا و مور او مص لحمصاد د ها ‎ty supper, portadady‏ تون اه موس ‎BE‏ ‏اه سوه هط سوه ره ما لول ۳ Bl Qore ceopketed ty kopkewect oct shoredtdtok or sharechortiens © Lockie ‏مها سوم رس رو ال با اس وا لو‎ ‎of coher processor.‏ ال مسجو روجا روج مكلجا يسما ص صا ‎© Occkecohereuy her ty be wotcktued — reads oud wrtes of data ta bP Per cet Pre kites! version of cha. ‎Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 موه‎ 

صفحه 17:
0 ‏واد‎ Ockerewy Protocol :ممصو عاص لوجاك عرذأ أمصصاصم برص امت یی و چم( 1# © ePore reudkry/umitey oo page, the page dst be locked tt shored/exchisive woe. © Octechiey 3 pace, the page wet be read Pro dk © @ePore uniochieg 0 paye, the page west be writes to disk Pt ues wodPied. © Qore cowplex protoonls wil Pewer dik reuds/urites exist. © Cocke wherewy protools Por shored-onthiey systews ore svar. Bock database paye i ‏سا و له‎ processor. Requests te Petck the paye or write tte disk are sect to the howe processor. Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 جوم‎ 

صفحه 18:
+ teireqery Prardobow 9 ‏سس ذا بحصي ملعك د خا مسحمرة)‎ oa wuliple processurehtehe! separa Por ‏هو رواب ولد‎ ۱ ‏روموت تا‎ Pore of isiraqery pordiciow : © Ieireopersion Pordebr — pardiebe the exevuioa of ruck ievicked operatic ihe ery. © ‏موسي‎ 6 ee ed Mie eee expressive i pared, the Pirst Porw soles beter wih tooreusiny pardheise bers ‏اه امه سا‎ tugles processed by cack operative te typicdly wore thao the cucber oP pperdicas toa query Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ea00 ©Sbervehnts, Cork ced Cnakershe 

صفحه 19:
+ Pardes! Provesetay oF Rektivad Operaiow © Ow dsrssiva of pardlel cots ‏تیه‎ ‎© reahoob queries: © shored-oohtey achieve ۰ ‏سم‎ P,, ‏و نت امه‎ P. و 11 © ‏مس للم رات لاه سا ماه ملسم‎ shared my shored-dsk systews. ۶ ‏اسب‎ Por shored-orhioy systews can thus be nuevo shored-wewory cand shoredtdsh spsiews. © Aowever, sowe vpicrizaicas way be possible. Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. a0 ©Sbervehnts, Cork ced Cnakershe 

صفحه 20:
يما معدا ‎do sorter.‏ اکن ك جو سار ۵ رن ‎ ‏تسس ماس سنا ‏« را جوم ال ۲ ‎۱ ‏مه ,اه و نت ی موم ین‎ sonic ott ies ‏و‎ ‎© dpe that he inthe romp are seul io provessor P, ‎© Pptores the tuples treveived tewporariy oa dst D, ‎© Okie step remuires VO ocd cxemanicdion pverkead. Bl Cock provessor ۵ sorts ts portion of he rekon broly. ‎Bh Bock processors exentes suxwe pperdtioa (sor) in pordel wih oher processors, wih cy kieracioa wik he vers (chia pardietrw). ‎Bind werye opercion i tivid: romge-ponicoiy exeures thot, Por (ja, the hey vokes ft provessor (Pore dl fess thoa the key vduew io P, ‎ean ‎Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. 

صفحه 21:
+ @ordel Gort (Ova) ‎Oot Dore‏ تسه الم ‎Bl Qem noe he rektion kay dren been portivoed wren dks Dy. Do (bs‏ ‎wwhetever axxner).‏ ‎Bock processor P lordly sorts the skis oa dk D,‏ ۱ ‎BL Phe sorted nue oo euck processor ure hea weryed yet ke Piso sorted‏ ‎rir.‏ ‎Bl Pordiekze he wercien of sorted nine oe Polos!‏ صما جه ‎ke sored portions ot cack provessor P, ore rene portioned‏ © ‎processors Py oy Pag.‏ ‎Cock processor P, perPorws a were on he sireums ce they ore‏ © ‎received, 7 vet a okie sorted nin.‏ ‎Dhe sored nine oa processors Ppyiy Pg are cocextecrted to opt fe‏ © ‎Prd rend.‏ ‎Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 »مه‎ 

صفحه 22:
تسه مها BO Dhe fon operaioa requires pairs oP tuples ty be tested to see P they suttoPy ‏جما‎ ‎Jpn coerdiiza, ced P they do, the pair te added te the iota cup. © Pardee ckerthes otlewpt ty spit he pairs ty be tested over ‏اوه‎ ‎processors. (Buck processor theo oowputes pot of the iota lool. 198 ‏تاه اد‎ step, the resubs Pow puck processor coo be colerted toyether i produce the Pic resul. Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مومه‎ 

صفحه 23:
@ortiowed ota مه اد لو ‎is possible to portica ke two‏ رم ی اوه نو و ۲ میم انس ‎he pia boul‏ هجوت وی مس ‎the‏ ویو وه و ‎ond wae ann‏ ماس مود ‎he‏ اج ‎bet rand‏ ا Brand seach ‏او بو‎ hiy opartions, dewied ry, 7) « ‏ی‎ ae ۴ Oua we ciher rome portray or husk particu. Brand sonst be patioged oa her pia oirbues nan 2@), veknp he sae nee ‎Pucca.‏ اس ی او ‏,6 موم و ده ولج بر له مم ‎Be Bock processor Prd yen Bp Baw oF he stank‏ ‎Xl ‎wethods coc be used. ‎Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. 

صفحه 24:
@orticaed Iota (Ovc.) Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مهمه‎ 

صفحه 25:
(Pragqeecbou(Rep pas Upia ۳ ‏وود‎ ot posable Bor scene i ‏وم‎ ‎© exp, renin comin, ‏سرجه ماسج‎ < 9.0. © Cor pes were portiogiag is ot upplicuble, porulekzaica ‏را امه سا موی‎ Preqeoet oad replicate ‏سس‎ © Depicted va cent side ۲ ‏سب مسا مسرب سم میدن‎ © Ove of the retiioas, 5015 is pontticoed; ‏اور بو‎ techoique roa be wed. © Dhe vher rebaivs, 5, & repiooted umves ull the processors. © Processor Pea bbouly copter the joie oP wih of ‏مامز روه يحاص د خا‎ ‏جات‎ Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مومه‎ 

صفحه 26:
‘@ السو نی (a) Asymmetric (b) Fragment and replicate fragment and replicate 

صفحه 27:
+ 0 222 2 Bl Beverd mee! redices the sizes of the rettiows of pack processor. 8 rte partied ‏صما‎ o Parties rp, Fy oF yoie B portioned tty a7 partion, Cys jen cern xy ponticatay techoique way be used. here onvet be of beust ‏ب‎ * 0 provevsors. .., P. oan Pyar Py 0۳, ۰ ۰ © Lebel he processors oe 5 oor “mar se Poo 90۰ 9 ۵ ‏و هس رخا مج سا عمجم‎ order to-do 90, nie replied 7 Poo, Py Pgs hte ot rephxted ‏رو‎ ۵ On ap 7 Poa, © ‏مسجت روط‎ con be wed of rack processor P,, Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 جومه‎ 

صفحه 28:
+ 0 222 2 Bok versioae of Praceecbernbrephrde work wah aay pis ooaion, stare every ‏جاه‎ ‎fa r-co be tested wih every tiple ko ‏بح‎ ۲ Dendy kas ot hither pret thos poriiectes, stor por oP he rekiiocs (Por ‏مه‎ ‎Prexeveubanhreviouy) or bk rekaie (Sar qouerd Brangreurarnbrepioe) have be rephrcted. امس سم یواسم با موب پمپ سا له ‎i‏ اه سا رو ۱ ‎Cx, ow ot odd ond ris ane, ond dead) portioced,‏ © حاط جما مج لمجم ‎aorzss ol processors, ther thas reparticg‏ = سار اه Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مومه‎ 

صفحه 29:
Cote Cadel Woek=loia (Porciiebates porrtioced bach joka Bl Qe ance ote swoler fron rowd hereore ote chose or the bull rekon. BE ook Ruanton fy thes the iota otic ‏جايس‎ oP each ture fs = od cape thir tuple to oe oP the processors. Bl Bock processor P reuse tuples of or that are oa ter desk D, ord sends eck unle to he oppropriie processor based oa hock Rumrioa fy. Let =; dewote te ‏ماود‎ of rektioa oth ae ped to processor P, © 0s tps oF rektiog 5 ore reveived of the destoaiva processors, they ore ‏لاور‎ Pudker ustey camher husk ‏ی و لح و ان ری رش‎ the (0) ها ملس Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مومه‎ 

صفحه 30:
+ Cwted Pade! Lotro (Oru) Bl Ocer he tuples of shave bees detrbted, the larger rehtion ris recotrbuted ores thew processors ern the husk ‏پا مش‎ © bet, dewte the tuples oP rebates thot ore seu iy processor P, Behe rps are revewed ot he devintion processors, hey ore repartioced “unten he Pucetiva A, © (het oe the probe retaiog tp porttivged int the sequectal kask-pic okpriher). ۱ Bock processor P executes the bukd ond probe phases of ‏مجامج جما‎ chprthey po he bel pariows ‏تناس لع سر‎ rod stp produce a portion of the Prod rend of he ‏ماسجا‎ Bl Oot: Week-pin opikotzatoce can be opphed to be pardlel cae جما اه وه اس و لح سا مه نموه باس تلو ‎ep, the‏ © دادم ای ما رت اه او با له لو موی و ود رس ‎theo bark ic.‏ Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. 

صفحه 31:
‎loin‏ مور لسنوو() اطل۵<) ‎ ‎۲ ‏جوا‎ thot ‏ما‎ sis work ‏.)ددسم برط له ات نا امه ما ما اه‎ ‎0 there is oo fedex: oa joi otribute oP reticg mot each oF the portions of ‏ع مهار‎ ‏اجه ,لاه بای ما ات ,ولو مر ‎Ose‏ ‏ما اه رام ماه سا روص ‎© Gtk processor P where ‏مور اه ما و‎ = ty stored reads the tuples of rehticn = stored to OD, od rephowtes the tuples to every uber provessur P, ‎© Otte ead of this phose, relation = rephooted of dl ster thot store tuples oF ‏ع مهار‎ ‎© Gtk processor 0 perPorns ‏اه و سل له مه‎ netics = watts the ‏ع اه ماو‎ ‎Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مهمه‎ 

صفحه 32:
Olker (Relipod Operdipes Grkvka ‏مايه‎ ‎BP 0 bof be Pore 9, =v, where 9 ‏لحي صحاف جه جا‎ vu uch. (AP re ‏مسا بط و بلط‎ & pePorwed ua ski ‏موصي‎ BP 0 bof be Por <= y <= u (e., 0 ‏جا‎ arennp sekotira) onl he rekios koe bere ‏مت ساس سح‎ © Gelevioa is perPorwed of cack provessor whose portiicg pverkaps ust the 4a db her ‏ماه سا تم‎ is perPorwed to pordlel of oll the provessors. Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مومه‎ 

صفحه 33:
+ Ober Rrktvad Operciow (Ovu.) ممصا جاص 0 11 وا بو امسصم جلا خان سجصلات مج روا موی ۶ ‎oe they ore Rove carter soe.‏ كمد جد ماص ل ‎riko‏ > © Ceci dby partion ie tuples (woke ether roses or kusk= porary) ond ‏وا جرب سوم‎ leroy ot perk peters 8 Preevivn © Creevics wihou duphodte ‏موی ماو‎ be perPorwed os tuples ore recd ta Prog desk i parcel. © 112 ‏مجقصاصاح صتتصاصك‎ is required, coy oP the ubove duphrote eerie techoiques con be werd. Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مومه‎ 

صفحه 34:
Croup Oqgreqaiva © Conic the rebaica cathe yecupteg atributies oad thea cocpuie the cpp ec valves locally ot rack provessor. BE Oca ether cost oP ineePerriog ples dure partivaiey by pany ooxop ater ‏رم انا سای مقس وی‎ ۳ ‏ور اسف مشچ موی توا‎ (ua those pkey sre ‏ال مس‎ 0 + ‏نج‎ fo ‏ی و مره وم ات ری‎ provessor. له روا میا من لحم ع موی ما سا ‎Result oP‏ © ‎(P, to yet the Prod resul‏ سس ره نب وت لور موه © ewer tuples ceed to be sect to ver processors durtay portiizcicr. Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 جومه‎ 

صفحه 35:
4 Oost oP Pardls! Guchnioa oP Operuiow OP there ts ow shew io the portticgiog, oad there is 07 overhead due to the parcel evolvaiod, expevied speed-up wil be (sr BAP chew ond overheats oe dbo ty be token hip arco, he ‏ناميهلا دجما‎ pened ‏ام یس با و سیم‎ ‎Dea)‏ )در ی اس مس اس تا 9 ط ‏ مامت و سا با تاره و ‎© ‏مور و مرن سا و ما سر با با‎ P, ‏صم ع لو رنه عمجت و بل لو با با له ‎his‏ ۱ موی و لور ‎Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مومه‎ 

صفحه 36:
حاط ی + دمم لس © تا سم( و مه سین ۶ ‎re‏ ال كا وك ودج اس و عم سا سا مه ما وم وق ۶ ‎es ee ee‏ . tor 2۲ ۵ ۳ ۱ ۱ 9 ۲ ‏ابو - وس اه مس ۲ اسوبی‎ rg » @ad PS be usstqaed the cowputetion oP temp re © Govk of hese opercives coo execute it pordtel, ‏اه قح ره‎ ۷ ‏جملا جا ون‎ uext pperdicg eyes op itis copies Parte remus * Provided 0 pipetocuble pia evdativa dyoriher (e.g. tidexed vested bop pk) & wed Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مومه‎ 

صفحه 37:
حاص صا جلاتججم سوه رات طلوه ۱ ‎pordletow tp verP dl stave‏ ی( ۲ ۲ ‏ای ای ان خی(‎ oP provessors, but doer oot sod up well wil ore processors. Our recsve is thot pipelor chotos ‏را میتی اه بت و‎ ما عم و له لس تج ‎Coco pipetoe vperciors whick do oot produce‏ © له سوه وم له ۱ ‏جاصعانب صا تناد‎ oe pperdor's exevuiva ovet ts wud higher thac the vers. Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 جومه‎ 

صفحه 38:
۳۹ Tadspeated Padstow few, Bl tadepeudedt pardietow © Consider a prt of ‏اس‎ rq Pr DX rg Dd rep Let P, be copied the cocputaion of 1 568 ١١ 9-1 000 ‏موس سا لو صا‎ oP ee = ro Re ١ Bud @, be wokned he erwp ction ‏السرم خات‎ ١ Py aed ‏و رخ‎ work tedepeudeddy to pardlet ©, bee tb wait Por ep Prow ®, od P, ~ One pipekee ouput ob Py ond Pt Po, coxobroisny kndepeodect ‏ماس‎ ‎ord pipetoed‏ طلسم ‎© Ove wt provide a high degre of pardlctsw ۱ ‏نها ه خن لخاسی‎ degree oP ‏ام‎ ‎* lees useful eo highly pardlel systew, ‎eae ‎Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. 

صفحه 39:
Query Opiwirdiva © Qhey opiezaivg to pordel databases te stqaPicadly wore ‏ما وی‎ query vpitrizaiog tr sequecital databases. ‎See ence ae eres‏ د نس يه ل ‎wosts ond issues suck os shew ood resvurce pooieaic.‏ ‎© Okeu scheduleg exevutivg tree fo parle systew, wet decide? © bow te pordielize ruck ppercica ced how woey provessvrs te use Por i. ‎© Oho operciccs to pipeter, whol operatives to execu ideprudediy it ‏له الوم‎ whol pperdicas tv exemue sequecichs, vor Per the vier. ‎© Oeterninteg he aout oP ‏موی وا ماه و تم‎ operon te 0 problew. ‎processors thee oped core result ist highs‏ سوه اه وق و اوه مش وین رو بت بط ‎Lowy pipetoes should be avoided us the Pied operatic way wat a‏ 18 ‎previous resvurces‏ لامها ‎hte‏ ‎ ‎Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. e186 ©Sbervehnts, Cork ced Cnakershe 

صفحه 40:
رن تاه هر ‎Gravy‏ + ی ان ‎the cxeober oP sequectd evokvatioa‏ © DherePore hewtotos ore ceeded while opivtzatira BP ive ohercave ‏مها‎ Por chovetey porcilel phos: © We pipebutcy aud ‏موس‎ pipehta) Net pordiekae every operuioa aoriss oll provessurs. © Pretax best phe ae cur ester = use stoscard ‏فا ,جنس مرت‎ uals cow woot seeded ” Okan pardlel ‏من سول‎ the echoage-opersior wd! ‏ما تحص ۵ بو تنم‎ query pee Wp portoa od detrbute tuples ~ eek pperaiva works tedepeudedly va lycud chta oa eack propessor, ta parle wuts viker copies ‏من سا ان‎ © Pret chovee covet ef Pied sequectd pho od thea hover how best pardiekze the ‏مرش رن‎ ter teat ‏.مهام‎ * Coa explore pipetoed pardleksw as aa opi BL Okoeorny a pod physi or gstzuirn (partizorny lechokne) ۰ Kopp tect epee ‏بف‎ ‏ی‎ Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مهمه‎ 

صفحه 41:
+ Dvstqa oP Pardel Opeewe (Gove tesues to the design oP parclel systews! © (Parddel Ioadiog of dota Brow exterodl sources ts oeeded fo order ty horde harce ‏مكلك مرس اه ماو‎ © ‏سا نا مایت‎ of sowe ‏معا مه و‎ © Probubliy oP soe disk or provessor Pater is higher ito parle ‏وه‎ © Operctos (perkops wil: dearaded perPorcmaace) should be promi te oper of Poke. © Redvadkrony uchieved by storey extra copy oP every dots tec of crazier provessur. Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. eee ©Sbervehnts, Cork ced Cnakershe 

صفحه 42:
+ Ovetqa oP Pade! Opstrwe (Ovu.) ۲ Ocrke reorprizaion of cht ond schewa champs cnet be supporied. ها ول موی سل سره وه مه لت ‎Cor exnople,‏ © ویو ‎days eyed ooo pordhel‏ ١ Deed i dlow oer process (Itserioce/tebices|arkiew) to be perPorwed oa rektioa ved oe inex tr bein ‏یو‎ © weir dew! index cousirurtivg tracks ‏کل ی له ماه‎ vo chores ot the eo. Bb ceed spp Bor ore repantionkny ad achewa choses (exerted oma nreddy wth ober provesskni)- Oxsdrer Gyetrw Oneewpe -O* Crim, Gui O, OOOO. ‏سا0 لح 0 لا سواه 1 مجمه‎ 

صفحه 43:
Gad oP Okaper 

Chapter 21: Parallel Databases Database System Concepts, 5th Ed. ©Silberschatz, Korth and Sudarshan See www.db-book.com for conditions on re-use Chapter 21: Parallel Databases  Introduction  I/O Parallelism  Interquery Parallelism  Intraquery Parallelism  Intraoperation Parallelism  Interoperation Parallelism  Design of Parallel Systems Database System Concepts - 5th Edition, Aug 22, 2005. 21.2 ©Silberschatz, Korth and Sudarshan Introduction    Parallel machines are becoming quite common and affordable  Prices of microprocessors, memory and disks have dropped sharply  Recent desktop computers feature multiple processors and this trend is projected to accelerate Databases are growing increasingly large  large volumes of transaction data are collected and stored for later analysis.  multimedia objects like images are increasingly stored in databases Large-scale parallel database systems increasingly used for:  storing large volumes of data  processing time-consuming decision-support queries  providing high throughput for transaction processing Database System Concepts - 5th Edition, Aug 22, 2005. 21.3 ©Silberschatz, Korth and Sudarshan Parallelism in Databases  Data can be partitioned across multiple disks for parallel I/O.  Individual relational operations (e.g., sort, join, aggregation) can be executed in parallel   data can be partitioned and each processor can work independently on its own partition. Queries are expressed in high level language (SQL, translated to relational algebra)  makes parallelization easier.  Different queries can be run in parallel with each other. control takes care of conflicts.  Thus, databases naturally lend themselves to parallelism. Database System Concepts - 5th Edition, Aug 22, 2005. 21.4 Concurrency ©Silberschatz, Korth and Sudarshan I/O Parallelism  Reduce the time required to retrieve relations from disk by partitioning  the relations on multiple disks.  Horizontal partitioning – tuples of a relation are divided among many disks such that each tuple resides on one disk.  Partitioning techniques (number of disks = n): Round-robin: Send the ith tuple inserted in the relation to disk i mod n. Hash partitioning:    Choose one or more attributes as the partitioning attributes. Choose hash function h with range 0…n - 1 Let i denote result of hash function h applied to tuple. Send tuple to disk i. Database System Concepts - 5th Edition, Aug 22, 2005. 21.5 the partitioning attribute value of a ©Silberschatz, Korth and Sudarshan I/O Parallelism (Cont.)  Partitioning techniques (cont.):  Range partitioning:  Choose an attribute as the partitioning attribute.  A partitioning vector [vo, v1, ..., vn-2] is chosen.  Let v be the partitioning attribute value of a tuple. Tuples such that vi  vi+1 go to disk I + 1. Tuples with v < v0 go to disk 0 and tuples with v  vn-2 go to disk n-1. E.g., with a partitioning vector [5,11], a tuple with partitioning attribute value of 2 will go to disk 0, a tuple with value 8 will go to disk 1, while a tuple with value 20 will go to disk2. Database System Concepts - 5th Edition, Aug 22, 2005. 21.6 ©Silberschatz, Korth and Sudarshan Comparison of Partitioning Techniques  Evaluate how well partitioning techniques support the following types of data access: 1.Scanning the entire relation. 2.Locating a tuple associatively – point queries.  E.g., r.A = 25. 3.Locating all tuples such that the value of a given attribute lies within a specified range – range queries.  E.g., 10  r.A < 25. Database System Concepts - 5th Edition, Aug 22, 2005. 21.7 ©Silberschatz, Korth and Sudarshan Comparison of Partitioning Techniques (Cont.) Round robin:  Advantages    Best suited for sequential scan of entire relation on each query. All disks have almost an equal number of tuples; retrieval work is thus well balanced between disks. Range queries are difficult to process  No clustering -- tuples are scattered across all disks Database System Concepts - 5th Edition, Aug 22, 2005. 21.8 ©Silberschatz, Korth and Sudarshan Comparison of Partitioning Techniques(Cont.) Hash partitioning:    Good for sequential access  Assuming hash function is good, and partitioning attributes form a key, tuples will be equally distributed between disks  Retrieval work is then well balanced between disks. Good for point queries on partitioning attribute  Can lookup single disk, leaving others available for answering other queries.  Index on partitioning attribute can be local to disk, making lookup and update more efficient No clustering, so difficult to answer range queries Database System Concepts - 5th Edition, Aug 22, 2005. 21.9 ©Silberschatz, Korth and Sudarshan Comparison of Partitioning Techniques (Cont.)  Range partitioning:  Provides data clustering by partitioning attribute value.  Good for sequential access  Good for point queries on partitioning attribute: only one disk needs to be accessed.  For range queries on partitioning attribute, one to a few disks may need to be accessed  Remaining disks are available for other queries.  Good if result tuples are from one to a few blocks.  If many blocks are to be fetched, they are still fetched from one to a few disks, and potential parallelism in disk access is wasted  Example of execution skew. Database System Concepts - 5th Edition, Aug 22, 2005. 21.10 ©Silberschatz, Korth and Sudarshan Partitioning a Relation across Disks  If a relation contains only a few tuples which will fit into a single disk block, then assign the relation to a single disk.  Large relations are preferably partitioned across all the available disks.  If a relation consists of m disk blocks and there are n disks available in the system, then the relation should be allocated min(m,n) disks. Database System Concepts - 5th Edition, Aug 22, 2005. 21.11 ©Silberschatz, Korth and Sudarshan Handling of Skew  The distribution of tuples to disks may be skewed — that is, some disks have many tuples, while others may have fewer tuples.  Types of skew:   Attribute-value skew.  Some values appear in the partitioning attributes of many tuples; all the tuples with the same value for the partitioning attribute end up in the same partition.  Can occur with range-partitioning and hash-partitioning. Partition skew.  With range-partitioning, badly chosen partition vector may assign too many tuples to some partitions and too few to others.  Less likely with hash-partitioning if a good hash-function is chosen. Database System Concepts - 5th Edition, Aug 22, 2005. 21.12 ©Silberschatz, Korth and Sudarshan Handling Skew in Range-Partitioning  To create a balanced partitioning vector (assuming partitioning attribute forms a key of the relation):  Sort the relation on the partitioning attribute.  Construct the partition vector by scanning the relation in sorted order as follows.   After every 1/nth of the relation has been read, the value of the partitioning attribute of the next tuple is added to the partition vector.  n denotes the number of partitions to be constructed.  Duplicate entries or imbalances can result if duplicates are present in partitioning attributes. Alternative technique based on histograms used in practice Database System Concepts - 5th Edition, Aug 22, 2005. 21.13 ©Silberschatz, Korth and Sudarshan Handling Skew using Histograms  Balanced partitioning vector can be constructed from histogram in a relatively straightforward fashion   Assume uniform distribution within each range of the histogram Histogram can be constructed by scanning relation, or sampling (blocks containing) tuples of the relation Database System Concepts - 5th Edition, Aug 22, 2005. 21.14 ©Silberschatz, Korth and Sudarshan Handling Skew Using Virtual Processor Partitioning   Skew in range partitioning can be handled elegantly using virtual processor partitioning:  create a large number of partitions (say 10 to 20 times the number of processors)  Assign virtual processors to partitions either in round-robin fashion or based on estimated cost of processing each virtual partition Basic idea:  If any normal partition would have been skewed, it is very likely the skew is spread over a number of virtual partitions  Skewed virtual partitions get spread across a number of processors, so work gets distributed evenly! Database System Concepts - 5th Edition, Aug 22, 2005. 21.15 ©Silberschatz, Korth and Sudarshan Interquery Parallelism  Queries/transactions execute in parallel with one another.  Increases transaction throughput; used primarily to scale up a transaction processing system to support a larger number of transactions per second.  Easiest form of parallelism to support, particularly in a shared-memory parallel database, because even sequential database systems support concurrent processing.  More complicated to implement on shared-disk or shared-nothing architectures  Locking and logging must be coordinated by passing messages between processors.  Data in a local buffer may have been updated at another processor.  Cache-coherency has to be maintained — reads and writes of data in buffer must find latest version of data. Database System Concepts - 5th Edition, Aug 22, 2005. 21.16 ©Silberschatz, Korth and Sudarshan Cache Coherency Protocol  Example of a cache coherency protocol for shared disk systems:  Before reading/writing to a page, the page must be locked in shared/exclusive mode.  On locking a page, the page must be read from disk  Before unlocking a page, the page must be written to disk if it was modified.  More complex protocols with fewer disk reads/writes exist.  Cache coherency protocols for shared-nothing systems are similar. Each database page is assigned a home processor. Requests to fetch the page or write it to disk are sent to the home processor. Database System Concepts - 5th Edition, Aug 22, 2005. 21.17 ©Silberschatz, Korth and Sudarshan Intraquery Parallelism  Execution of a single query in parallel on multiple processors/disks; important for speeding up long-running queries.  Two complementary forms of intraquery parallelism :  Intraoperation Parallelism – parallelize the execution of each individual operation in the query.  Interoperation Parallelism – execute the different operations in a query expression in parallel. the first form scales better with increasing parallelism because the number of tuples processed by each operation is typically more than the number of operations in a query Database System Concepts - 5th Edition, Aug 22, 2005. 21.18 ©Silberschatz, Korth and Sudarshan Parallel Processing of Relational Operations  Our discussion of parallel algorithms assumes:  read-only queries  shared-nothing architecture  n processors, P0, ..., Pn-1, and n disks D0, ..., Dn-1, where disk Di is associated with processor Pi.  If a processor has multiple disks they can simply simulate a single disk Di.  Shared-nothing architectures can be efficiently simulated on shared-memory and shared-disk systems.  Algorithms for shared-nothing systems can thus be run on shared-memory and shared-disk systems.  However, some optimizations may be possible. Database System Concepts - 5th Edition, Aug 22, 2005. 21.19 ©Silberschatz, Korth and Sudarshan Parallel Sort Range-Partitioning Sort  Choose processors P0, ..., Pm, where m  n -1 to do sorting.  Create range-partition vector with m entries, on the sorting attributes  Redistribute the relation using range partitioning  all tuples that lie in the ith range are sent to processor Pi  Pi stores the tuples it received temporarily on disk Di.  This step requires I/O and communication overhead.  Each processor Pi sorts its partition of the relation locally.  Each processors executes same operation (sort) in parallel with other processors, without any interaction with the others (data parallelism).  Final merge operation is trivial: range-partitioning ensures that, for 1 j m, the key values in processor Pi are all less than the key values in Pj. Database System Concepts - 5th Edition, Aug 22, 2005. 21.20 ©Silberschatz, Korth and Sudarshan Parallel Sort (Cont.) Parallel External Sort-Merge  Assume the relation has already been partitioned among disks D0, ..., Dn-1 (in whatever manner).  Each processor Pi locally sorts the data on disk Di.  The sorted runs on each processor are then merged to get the final sorted output.  Parallelize the merging of sorted runs as follows:  The sorted partitions at each processor Pi are range-partitioned across the processors P0, ..., Pm-1.  Each processor Pi performs a merge on the streams as they are received, to get a single sorted run.  The sorted runs on processors P0,..., Pm-1 are concatenated to get the final result. Database System Concepts - 5th Edition, Aug 22, 2005. 21.21 ©Silberschatz, Korth and Sudarshan Parallel Join  The join operation requires pairs of tuples to be tested to see if they satisfy the join condition, and if they do, the pair is added to the join output.  Parallel join algorithms attempt to split the pairs to be tested over several processors. Each processor then computes part of the join locally.  In a final step, the results from each processor can be collected together to produce the final result. Database System Concepts - 5th Edition, Aug 22, 2005. 21.22 ©Silberschatz, Korth and Sudarshan Partitioned Join  For equi-joins and natural joins, it is possible to partition the two input relations across the processors, and compute the join locally at each processor.  Let r and s be the input relations, and we want to compute r  r and s each are partitioned into n partitions, denoted r0, r1, ..., rn-1 and s0, s1, ..., sn-1.  Can use either range partitioning or hash partitioning.  r and s must be partitioned on their join attributes r.A and s.B), using the same rangepartitioning vector or hash function.  Partitions ri and si are sent to processor Pi,  Each processor Pi locally computes ri methods can be used. Database System Concepts - 5th Edition, Aug 22, 2005. 21.23 ri.A=si.B si. r.A=s.B s. Any of the standard join ©Silberschatz, Korth and Sudarshan Partitioned Join (Cont.) Database System Concepts - 5th Edition, Aug 22, 2005. 21.24 ©Silberschatz, Korth and Sudarshan Fragment-and-Replicate Join  Partitioning not possible for some join conditions   For joins were partitioning is not applicable, parallelization can be accomplished by fragment and replicate technique   e.g., non-equijoin conditions, such as r.A > s.B. Depicted on next slide Special case – asymmetric fragment-and-replicate:  One of the relations, say r, is partitioned; any partitioning technique can be used.  The other relation, s, is replicated across all the processors.  Processor Pi then locally computes the join of ri with all of s using any join technique. Database System Concepts - 5th Edition, Aug 22, 2005. 21.25 ©Silberschatz, Korth and Sudarshan Depiction of Fragment-and-Replicate Joins Database System Concepts - 5th Edition, Aug 22, 2005. 21.26 ©Silberschatz, Korth and Sudarshan Fragment-and-Replicate Join (Cont.)  General case: reduces the sizes of the relations at each processor.  r is partitioned into n partitions,r0, r1, ..., r n-1;s is partitioned into m partitions, s0, s1, ..., sm-1.  Any partitioning technique may be used.  There must be at least m * n processors.  Label the processors as  P0,0, P0,1, ..., P0,m-1, P1,0, ..., Pn-1m-1.  Pi,j computes the join of ri with sj. In order to do so, ri is replicated to Pi,0, Pi,1, ..., Pi,m-1, while si is replicated to P0,i, P1,i, ..., Pn-1,i  Any join technique can be used at each processor Pi,j. Database System Concepts - 5th Edition, Aug 22, 2005. 21.27 ©Silberschatz, Korth and Sudarshan Fragment-and-Replicate Join (Cont.)  Both versions of fragment-and-replicate work with any join condition, since every tuple in r can be tested with every tuple in s.  Usually has a higher cost than partitioning, since one of the relations (for asymmetric fragment-and-replicate) or both relations (for general fragment-and-replicate) have to be replicated.  Sometimes asymmetric fragment-and-replicate is preferable even though partitioning could be used.  E.g., say s is small and r is large, and already partitioned. It may be cheaper to replicate s across all processors, rather than repartition r and s on the join attributes. Database System Concepts - 5th Edition, Aug 22, 2005. 21.28 ©Silberschatz, Korth and Sudarshan Partitioned Parallel Hash-Join Parallelizing partitioned hash join:  Assume s is smaller than r and therefore s is chosen as the build relation.  A hash function h1 takes the join attribute value of each tuple in s and maps this tuple to one of the n processors.  Each processor Pi reads the tuples of s that are on its disk Di, and sends each tuple to the appropriate processor based on hash function h1. Let si denote the tuples of relation s that are sent to processor Pi.  As tuples of relation s are received at the destination processors, they are partitioned further using another hash function, h2, which is used to compute the hash-join locally. (Cont.) Database System Concepts - 5th Edition, Aug 22, 2005. 21.29 ©Silberschatz, Korth and Sudarshan Partitioned Parallel Hash-Join (Cont.)  Once the tuples of s have been distributed, the larger relation r is redistributed across the m processors using the hash function h1   Let ri denote the tuples of relation r that are sent to processor Pi. As the r tuples are received at the destination processors, they are repartitioned using the function h2  (just as the probe relation is partitioned in the sequential hash-join algorithm).  Each processor Pi executes the build and probe phases of the hash-join algorithm on the local partitions ri and s of r and s to produce a partition of the final result of the hash-join.  Note: Hash-join optimizations can be applied to the parallel case  e.g., the hybrid hash-join algorithm can be used to cache some of the incoming tuples in memory and avoid the cost of writing them and reading them back in. Database System Concepts - 5th Edition, Aug 22, 2005. 21.30 ©Silberschatz, Korth and Sudarshan Parallel Nested-Loop Join  Assume that  relation s is much smaller than relation r and that r is stored by partitioning.  there is an index on a join attribute of relation r at each of the partitions of relation r.  Use asymmetric fragment-and-replicate, with relation s being replicated, and using the existing partitioning of relation r.  Each processor Pj where a partition of relation s is stored reads the tuples of relation s stored in Dj, and replicates the tuples to every other processor Pi.   At the end of this phase, relation s is replicated at all sites that store tuples of relation r. Each processor Pi performs an indexed nested-loop join of relation s with the ith partition of relation r. Database System Concepts - 5th Edition, Aug 22, 2005. 21.31 ©Silberschatz, Korth and Sudarshan Other Relational Operations Selection (r)  If  is of the form ai = v, where ai is an attribute and v a value.   If  is of the form l <= ai <= u (i.e.,  is a range selection) and the relation has been range-partitioned on ai   If r is partitioned on ai the selection is performed at a single processor. Selection is performed at each processor whose partition overlaps with the specified range of values. In all other cases: the selection is performed in parallel at all the processors. Database System Concepts - 5th Edition, Aug 22, 2005. 21.32 ©Silberschatz, Korth and Sudarshan Other Relational Operations (Cont.)  Duplicate elimination  Perform by using either of the parallel sort techniques    eliminate duplicates as soon as they are found during sorting. Can also partition the tuples (using either range- or hash- partitioning) and perform duplicate elimination locally at each processor. Projection  Projection without duplicate elimination can be performed as tuples are read in from disk in parallel.  If duplicate elimination is required, any of the above duplicate elimination techniques can be used. Database System Concepts - 5th Edition, Aug 22, 2005. 21.33 ©Silberschatz, Korth and Sudarshan Grouping/Aggregation  Partition the relation on the grouping attributes and then compute the aggregate values locally at each processor.  Can reduce cost of transferring tuples during partitioning by partly computing aggregate values before partitioning.  Consider the sum aggregation operation:  Perform aggregation operation at each processor Pi on those tuples stored on disk Di    results in tuples with partial sums at each processor. Result of the local aggregation is partitioned on the grouping attributes, and the aggregation performed again at each processor Pi to get the final result. Fewer tuples need to be sent to other processors during partitioning. Database System Concepts - 5th Edition, Aug 22, 2005. 21.34 ©Silberschatz, Korth and Sudarshan Cost of Parallel Evaluation of Operations  If there is no skew in the partitioning, and there is no overhead due to the parallel evaluation, expected speed-up will be 1/n  If skew and overheads are also to be taken into account, the time taken by a parallel operation can be estimated as Tpart + Tasm + max (T0, T1, …, Tn-1)  Tpart is the time for partitioning the relations  Tasm is the time for assembling the results  Ti is the time taken for the operation at processor Pi  this needs to be estimated taking into account the skew, and the time wasted in contentions. Database System Concepts - 5th Edition, Aug 22, 2005. 21.35 ©Silberschatz, Korth and Sudarshan Interoperator Parallelism  Pipelined parallelism  Consider a join of four relations    r1 r2 r3 r4 Set up a pipeline that computes the three joins in parallel  Let P1 be assigned the computation of temp1 = r1 r2  And P2 be assigned the computation of temp2 = temp1  And P3 be assigned the computation of temp2 r3 r4 Each of these operations can execute in parallel, sending result tuples it computes to the next operation even as it is computing further results  Provided a pipelineable join evaluation algorithm (e.g. indexed nested loops join) is used Database System Concepts - 5th Edition, Aug 22, 2005. 21.36 ©Silberschatz, Korth and Sudarshan Factors Limiting Utility of Pipeline Parallelism  Pipeline parallelism is useful since it avoids writing intermediate results to disk  Useful with small number of processors, but does not scale up well with more processors. One reason is that pipeline chains do not attain sufficient length.  Cannot pipeline operators which do not produce output until all accessed (e.g. aggregate and sort)  Little speedup is obtained for the frequent cases of skew in which operator's execution cost is much higher than the others. Database System Concepts - 5th Edition, Aug 22, 2005. 21.37 inputs have been one ©Silberschatz, Korth and Sudarshan Independent Parallelism  Independent parallelism  Consider a join of four relations r1 r2 r3 r4  Let P1 be assigned the computation of temp1 = r1 r2  And P2 be assigned the computation of temp2 = r3  And P3 be assigned the computation of temp1  P1 and P2 can work independently in parallel  P3 has to wait for input from P1 and P2 r4 temp2 – Can pipeline output of P1 and P2 to P3, combining independent parallelism and pipelined parallelism  Does not provide a high degree of parallelism  useful with a lower degree of parallelism.  less useful in a highly parallel system, Database System Concepts - 5th Edition, Aug 22, 2005. 21.38 ©Silberschatz, Korth and Sudarshan Query Optimization  Query optimization in parallel databases is significantly more complex than query optimization in sequential databases.  Cost models are more complicated, since we must take into account partitioning costs and issues such as skew and resource contention.  When scheduling execution tree in parallel system, must decide:   How to parallelize each operation and how many processors to use for it.  What operations to pipeline, what operations to execute independently in parallel, and what operations to execute sequentially, one after the other. Determining the amount of resources to allocate for each operation is a problem.   E.g., allocating more processors than optimal can result in high communication overhead. Long pipelines should be avoided as the final operation may wait a lot for inputs, while holding precious resources Database System Concepts - 5th Edition, Aug 22, 2005. 21.39 ©Silberschatz, Korth and Sudarshan Query Optimization (Cont.)  The number of parallel evaluation plans from which to choose from is much larger than the number of sequential evaluation plans.   Therefore heuristics are needed while optimization Two alternative heuristics for choosing parallel plans:  No pipelining and inter-operation pipelining; just parallelize every operation across all processors.  Finding best plan is now much easier --- use standard optimization technique, but with new cost model  Volcano parallel database popularize the exchange-operator model – exchange operator is introduced into query plans to partition and distribute tuples – each operation works independently on local data on each processor, in parallel with other copies of the operation  First choose most efficient sequential plan and then choose how best to parallelize the operations in that plan.   Can explore pipelined parallelism as an option Choosing a good physical organization (partitioning technique) is important to speed up queries. Database System Concepts - 5th Edition, Aug 22, 2005. 21.40 ©Silberschatz, Korth and Sudarshan Design of Parallel Systems Some issues in the design of parallel systems:  Parallel loading of data from external sources is needed in order to handle large volumes of incoming data.  Resilience to failure of some processors or disks.  Probability of some disk or processor failing is higher in a parallel system.  Operation (perhaps with degraded performance) should be possible in spite of failure.  Redundancy achieved by storing extra copy of every data item at another processor. Database System Concepts - 5th Edition, Aug 22, 2005. 21.41 ©Silberschatz, Korth and Sudarshan Design of Parallel Systems (Cont.)  On-line reorganization of data and schema changes must be supported.  For example, index construction on terabyte databases can take hours or days even on a parallel system.    Need to allow other processing (insertions/deletions/updates) to be performed on relation even as index is being constructed. Basic idea: index construction tracks changes and ``catches up'‘ on changes at the end. Also need support for on-line repartitioning and schema changes (executed concurrently with other processing). Database System Concepts - 5th Edition, Aug 22, 2005. 21.42 ©Silberschatz, Korth and Sudarshan End of Chapter Database System Concepts, 5th Ed. ©Silberschatz, Korth and Sudarshan See www.db-book.com for conditions on re-use