Composite Events for Active Databases: Semantics, S. Chakravarthy V. Krishnaprasad E. Anwar S.-K. Kim

Transcription

1 Composit Evnts for Activ Databass: Smantics, Contxts and Dtction S. Chakravarthy V. Krishnaprasad E. Anwar S.-K. Kim Databas Systms Rsarch and Dvlopmnt Cntr Computr and Information Scincs Dpartmnt Univrsity of Florida, Gainsvill, FL 6 mail: fsharma, vk, msa, skkg@cis.ufl.du Abstract Making a databas systm activ ntails dvloping an xprssiv vnt spcication languag with wll-dnd smantics, algorithms for th dtction of composit vnts, and an architctur for an vnt dtctor along with its implmntation. This papr prsnts th smantics of composit vnts using th notion of a global vnt history (or a global vnt-log). Paramtr contxts ar introducd and prcisly dnd to facilitat cint managmnt and dtction of composit vnts. Finally,anarchitctur and th implmntation of a composit vnt dtctor is analyzd in th contxt of an objct-orintd activ DBMS. Introduction This papr focuss on th vnt componnt of th ECA (vnt-condition-action) ruls usd in activ databass. An ECA rul consists, primarily, of thr componnts: an vnt, a condition, and an action. A signicant body of work xists on ruls and rul procssing in a DBMS. Howvr, th vnt componnt of ruls has rcivd attntion only rcntly [Mis9, CM94, GJS9b, GD9] and prhaps is th last undrstood compard to th condition and action componnts. Conditions and actions corrspond to sid-ct fr quris and transactions, rspctivly. Although vnt spcication has bn addrssd in th litratur primarily in th contxt of activ Prmission to copy without f allorpart of this matrial is grantd providd that th copis ar not mad or distributd for dirct commrcial advantag, th VLDB copyright notic and th titl of th publication and its dat appar, and notic is givn that copying is by prmission of th Vry Larg Data Bas Endowmnt. To copy othrwis, or to rpublish, rquirs a f and/or spcial prmission from th Endowmnt. Procdings of th 0th VLDB Confrnc Santiago, Chil, 994 databass, its applicability is not limitd to activ databass. An xprssiv vnt spcication languag and its dtction can b usd for analyzing vnt historis (or vnt logs) [SW9] in applications, such asstock trading, trnd/dmographic prol computation, and auditing (ithr as vnts occur or ovr stord vnt occurrncs). Som aspcts of knowldg discovry (.g., dtrmining vnts that lad to th stock markt crash, undrstanding squncs of vnts lading to an arthquak) involv analyzing vnt pattrns and thir ct on various rcordd obsrvations. In othr words, applications that xamin caus-ct rlationships nd to spcify and dtct complx vnt pattrns. From th abov, it is vidnt that support for ruls nds to b complmntd with an xprssiv vnt spcication languag. As an xampl, managmnt of portfolios for various customrs in a stock-trading application may rquir a rul of th form \whn th DowJons avrag changs by 0%inany hour intrval, aftr raching th valu 750, xcut Trump's portfolio modl to dtrmin what to buy or sll". This rul rquirs not only th databas vnts, but also tmporal and composit vnts this rul rquirs an xprssiv vnt spcication languag for modling its vnts. As anothr xampl, dtcting apriodic occurrncs of a pattrn of vnts may indicat a potntial mony laundring schm. This papr xtnds arlir work on Snoop [Mis9, CM94] in svral signicant ways. Earlir work was primarily concrnd with th motivation for th vnt languag, classication of vnts, nd for vnt oprators, and th st of vnt oprators. In this papr, w introduc primitiv vnt squncs as ordrd occurrncs of a primitiv vnt (trmd primitiv vnt history/vnt-log), and composit vnt history/vntlog as a partial ordr of th mrgd primitiv (or othr composit) vnt historis. W dn th smantics of primitiv and composit vnts ovr an vnt history. W argu that th dtction of composit vnts ovr a composit vnt history lads to monotonically incrasing storag ovrhad as prvious occurrncs of vnts

2 cannot b dltd. To ovrcom this problm, w introduc th notion of paramtr contxts as a mchanism for prcisly rstricting th occurrncs that mak a composit vnt occur as wll as for computing its paramtrs. W hav dvlopd complt algorithms for dtcting Snoop xprssions in all paramtr contxts. Od [GJS9b, GJS9a] and Samos [GD9, GD94] addrss vnt spcication and dtction in th contxt of activ databass. Although thr ar som dirncs btwn Snoop, Od, and Samos in th vnt spcication languag (for xampl, Samos has a Tims oprator for dning th occurrnc of n vnts in an intrval and Od has a complmnt oprator), thy dir primarily in th mchanism usd for vnt dtction. Od uss a nit automaton and Samos uss a labld Ptri Nt. In contrast, w us an vnt graph whr vnt occurrncs (both primitiv and composit) ow bottom-up from th nods to thir parnts. Th formulation of vnt historis as prsntd in this papr is dirnt from that of Od [GJS9a]. In addition, paramtr contxts as wll as vnt dtction in various paramtr contxts ar xtnsions of our arlir work. Th rst of this papr is structurd as follows. Sction provids smantics for th oprators of Snoop [Mis9, CM94]. In sction, w dn global, primitiv, and composit vnt historis and prsnt th computation of composit vnts using th vnt history. In sction 4, w dn paramtr contxts and illustrat vnt dtction for all contxts. Sction 5 provids an architctur for composit vnt dtction in an objctorintd DBMS and highlights implmntation choics. Sction 6 contains conclusions. Smantics of Snoop For th purpos of this papr, w assum an qui-distant discrt tim domain having \0" as th origin and ach tim point rprsntd by a non-ngativ intgr. W distinguish btwn an vnt, an vnt xprssion, and an vnt modir. An vnt is dnd to b an instantanous, atomic (happns compltly or not at all) occurrnc of intrst at a point in tim. In databas applications, th intrst in vnts coms mostly from th stat changs that ar producd by data manipulation oprations. Similar vnts can b groupd into an vnt typ, and a typ of vnts can b furthr classid into subtyps, rsulting in an vnt typ hirarchy as th class hirarchy. For instanc, vnts of databas updat can b groupd into an vnt typ Updat, and furthr groupd into Updat-IBM, Updat-DEC, tc. Dirnt vnt typs ar distinguishd by dirnt vnt typ nams. Evnt of an vnt typ may occur zro or mor tims ovr th tim lin th tim of occurrnc of an vnt isdnotd by t occ. In addition, for simplicity, w assum Rtrival oprations may also b rgardd as vnts, although thy do not chang th databas stat. Or thy may b distinguishd by using paramtrs such as Updat(IBM) and Updat(DEC). that two occurrncs of th sam vnt typ ar not simultanous. Furthrmor, an vnt may causally prcd or follow anothr, or vnts may b unrlatd. For xampl, th twovnts nd-of-abort T and bgin-ofrollback Tmust follow on anothr and ar causally rlatd (causally dpndnt), whras th vnts bginof T and bgin-of T ar causally indpndnt and ar said to b unrlatd. An vnt isdnit if and only if it is guarantd to occur. An vnt typ is xprssd by anvnt xprssion, which shall b discussd in th following subsction. Although an vnt is assumd to instantanously occur at atimpoint, th vnt might b initiatd at a prior tim point, thus yilding a closd tim intrval btwn th start and nd points (t occ is th nd point by dfault). A transaction vnt issuch an xampl. In ordr to xplicitly spcify (or modify) th occurrnc tim of an vnt spanning a tim intrval, vnt modirs bgin-of and nd-of wr introducd in [CM94]. For inhrntly instantanous vnts, th two vnt modirs yild th sam tim point.. Primitiv Evnts Primitiv vnts ar thos that ar pr-dnd in th systm (and using th vnt modirs). A mchanism for th dtction is assumd to b availabl (rfr to [AMC9] for dtails). Primitiv vnts includ databas vnts, tmporal vnts, and xplicit vnts. Databas vnts corrspond to databas oprations, such as data manipulation oprations, transactions, or mthods in objct-orintd databass. Tmporal vnts ar ithr absolut or rlativ. An absolut tmporal vnt is spcid with an absolut valu of tim, and rprsntd as: < tim string > using th format < (hh/mm/ss)mm/dd/yy >. A rlativ tmporal vnt also corrsponds to a uniqu point on th tim lin but in this cas both th rfrnc point and th ost ar xplicitly spcid. Th rfrnc point may b any vnt that can b spcid in Snoop including an absolut tmporal vnt. Th syntax for a rlativ vnt isvnt +[tim string ]. Obsrv that th rlativ vnt subsums th absolut vnt. Howvr, th absolut vrsion is rtaind for practical rasons. Explicit vnts ar thos vnts that ar dtctd along with thir paramtrs by application programs (i.., outsid th DBMS) and ar only managd by th DBMS. Onc rgistrd with th systm, thy can b usd as primitiv vnts.. Evnt Exprssions and Evnt Oprators Primitiv vnts discussd so far ar usful for modling a numbr of applications. Howvr, for many othr applications, it is ncssary to dtct crtain combinations of dirnt vnts as a singl vnt, i.., a composit vnt. In this papr, a composit vnt is dnd by applying an vnt oprator to constitunt vnts that ar primitiv or othr composit vnts. In th absnc

3 of vnt oprators, svral ruls ar rquird to spcify a composit vnt. Furthrmor, som control information nds to b mad a part of a rul spcication. As mntiond bfor, an vnt typ is dnotd by an vnt xprssion. A primitivvnt (typ) nam itslf is an vnt xprssion. If E E E n ar vnt xprssions, an application of any vnt oprator, dscribd blow, ovr th vnt xprssions is an vnt xprssion. For an vnt E, bgin-of E, nd-of E, and (E) ar all vnt xprssions. If vnt modir is omittd, nd-of is assumd by dfault. Th oprator smantics dscribd blow assums th nd-of modir... Oprator Smantics An vnt E (ithr primitiv or composit) is a function from th tim domain onto th boolan valus, Tru and Fals. E : T!fTru Falsg givn by ( T(ru) if an vnt oftyp E occurs E(t) = at tim point t F(als) othrwis. W dnot th ngation of th boolan function E as E. Givn a tim point, it computs th nonoccurrnc of an vnt at that point. Th Snoop vnt oprators 4 and th smantics of composit vnts formd by ths vnt oprators ar as follows:. OR (r): Disjunction of two vnts E and E, dnotd E re, occurs whn E occurs or E occurs. Formally, (E re )(t) =E (t) _ E (t):. AND (4): Conjunction of two vnts E and E, dnotd E 4E, occurs whn both E and E occur, irrspctiv of thir ordr of occurrnc. Formally, (E 4E )(t) = (9t )(((E (t ) ^ E (t)) _(E (t ) ^ E (t))) ^ t t): Not that th OR and AND oprators ar commutativ and associativ: (E re )(t) =(E re )(t) ((E 4E )4E )(t) =(E 4(E 4E ))(t): In fact, in production rul systms (.g., OPS5 [For8, FM87]), programs ar writtn by incorporating a lot of control information as part of ruls which hav a form similar to an ECA rul. Spcically, in an OPS5 rul, vnts ar not xplicitly spcid but ar infrrd for th worst cas scnario. 4 W dnot th \disjunction", \conjunction", and \not" vnt oprators as r, 4, and :, rspctivly. Th symbols _, ^, and rprsnt th \or", \and", and \not" boolan oprators, rspctivly.. ANY: Th conjunction vnt, dnotd by ANY(m E E E n ) whr m n, occurs whn m vnts out of th n distinct vnts spcid occur, ignoring th rlativ ordr of thir occurrnc. Formally, ANY(m E E E n )(t) = (9t )(9t ) (9t m; ) (E i (t ) ^ E j (t ) ^^E k (t m; ) ^ E l (t) ^ (t t t m; t) ^ ( i j k l n) ^ (i 6= j 6= 6= k 6= l)): For xampl, ANY( E E E n )(t) =(9t )(9t ) (E i (t ) ^ E j (t ) ^ E k (t) ^ (t t t) ^ ( i j k n) ^ (i 6= j 6= k)): Also to spcify m distinct occurrncs of an vnt E, th following variant is providd: ANY(m E )(t) =(9t )(9t ) (9t m; ) (E(t ) ^ E(t ) ^^E(t m; ) ^ E(t) ^ (t <t < <t m; <t)): 4. SEQ ( ): Squnc of two vnts E and E,dnotd E E, occurs whn E occurs providd E has alrady occurrd. This implis that th tim of occurrnc of E is guarantd to b lss than th tim of occurrnc of E.Formally, (E E )(t) =(9t )(E (t ) ^ E (t) ^ (t <t)): It is possibl that aftr th occurrnc of E, E dos not occur at all. To avoid this situation, it is dsirabl that dnit vnts, such as nd-of-transaction or an absolut tmporal vnt, ar usd appropriatly. 5. Apriodic Oprators (A, A*): Th Apriodic oprator A allows on to xprss th occurrncs of an apriodic vnt within a closd tim intrval. Thr ar two vrsions of this vnt spcication. Th non-cumulativ apriodic vnt is xprssd as A(E E E ), whr E, E,andE ar arbitrary vnts. Th vnt A is signald ach tim E occurs within th tim intrval startd by E and ndd by E.Formally, A(E E E )(t) =(9t )(8t )(E (t ) ^ E (t) ^ (t t) ^ ((t t <t)!e (t ))): On th othr hand, th cumulativ apriodic vnt A (E E E ) occurs only onc whn E occurs and accumulats th occurrncs of E within th opn tim intrval formd by E and E. This constructor is usful for intgrity chcking in databass and for collcting paramtrs of an vnt ovr an intrval for

4 computing aggrgats. As an xampl, highst or lowst stock pric can b computd ovr an intrval using this oprator. Not that E itslf can occurs zro or mor tims within th intrval and dos not contribut to th occurrnc of th composit vnt A. Nonthlss, th paramtrs of A will contain th paramtrs of E.Formally, A (E E E )(t) = (9t )(E (t ) ^ E (t) ^ (t <t)): 6. Priodic Evnt Oprators (P, P*): A priodic vnt is a tmporal vnt that occurs priodically. A priodic vnt is dnotd as P(E TI [: paramtrs] E ) whr E and E ar vnts and TI [: paramtrs] is a tim intrval spcication with optional paramtr list. P occurs for vry TI intrval, starting aftr E and casing aftr E. Paramtrs spcid ar collctd ach timp occurs. If not spcid, th occurrnc tim of P is collctd by dfault. Th occurrnc of P is formally dnd as P(E TI [: paramtrs] E )(t) =(9t )(8t ) (E (t ) ^ ((t t t)!e (t )) ^ (t = t + i TI for som intgr i )): P has a cumulativ vrsion P* xprssd as P (E TI : paramtrs E ). Unlik P, P* occurs only onc whn E occurs. Also, spcid paramtrs ar collctd and accumulatd at th nd of ach priod and mad availabl whn P* occurs. Not that th paramtr spcication is mandatory in P*. Th occurrnc of P* is formally dnd as P (E TI : paramtrs E )(t) = (9t )(E (t ) ^ E (t) ^ (t t +TI)): 7. NOT (:): Th NOT oprator, dnotd :(E )[E E ], dtcts th non-occurrnc of th vnt E in th closd intrval formd by E and E. 5 Formally, :(E )[E E ](t) =(9t )(8t ) (E (t ) ^E (t) ^ E (t) ^ ((t t <t)!(e (t ) _ E (t )))):. Exampls Blow, w show som ruls that ntail dtction of various composit vnts. W us a simplid syntax to mak th vnts radabl.. Sampl IBM stock vry 0 minuts from 8 a.m. to 5 p.m. ach day. Evnt: P*(8 a.m., [0 mins]: IBMstock-pric, 5 p.m.). Whn 4 withdrawals ar mad on an account ina day, do not allow furthr withdrawals. Evnt: A(8 a.m., ANY(4, withdraw-on-an-account*), 5 p.m.) 5 Not that this oprator is dirnt from that of!e, a unary oprator in Od [GJS9b], which dtcts th occurrnc of any vnt othr than E.. In a nuclar powr plant ifthrisachang in ssion rat followd by an incras in tmpratur, introduc modrator rods. Evnt: (ssion-rat tmpratur-incras) 4. Comput th nw DowJons avrag whn any two of IBM, DEC, or Boing stock prics chang during th day. Evnt: A(8 a.m., ANY(, modify-ibm, modify-dec, modify-boing), 5 p.m.) Historis and Evnt Logs So far, w hav dnd th smantics of vnt oprators ovr th tim lin in which only th tim of (primitiv or composit) vnt occurrncs wr rcordd. Howvr, dtction of a composit vnt ntails dtcting not only th tim at which th composit vnt occurs, but also th spcic constitunt vnt occurrncs that mak th composit vnt occur. In this sction, w formally xprss th occurrnc of a composit vnt E with rspct to its constitunt vnts that form part of th occurrnc of E. Atsomlvl, th constitunt vnts ar primitivvnts. W dnot an occurrnc of an vnt typ E j by i j whr i indicats th rlativ tim of occurrnc with rspct to othr occurrncs of th sam vnt. Composit vnts ar rprsntd as a st of constitunt vnt occurrncs within which thordr of vnt occurrncs is prsrvd. Not that it is possibl for th sam constitunt vnt occurrnc to b usd for mor than on occurrnc of a composit vnt. Th last vnt inth st is on whos occurrnc mad th composit vnt occur. Th tim of occurrnc of a composit vnt is th tim of occurrnc of th last constitunt vnt. Global Evnt History/Evnt Log is a st of all primitiv vnt occurrncs and is dnotd by H. Each primitiv vnt occurrnc is rprsntd as a singlton stinthlog. H = ff i jgj for all j, primitiv vnt i j has occurrd at instanc i rlativ tovnts E j :g Primitiv Evnt History/Evnt Log of th primitiv vnt typ E is a st of th occurrncs of E prsnt in th Global History H and is dnotd by E[H]. E j [H]=ff i jgj for all i f i jghg: Composit Evnt History/Evnt Log ofacomposit vnt E that has n constitunt vnts E,, E n is a mapping from U th U global vnt history U H to a subst of E [H] E n [H] whr is an oprator that computs th cross product of two sts (whos lmnts ar sts) and mrgs th lmnts of th cross product using th union oprator. For xampl, givn vnt historis E [H] =ff 4 g f gg and E [H]=ff g f gg, E [H] U E [H] = ff 4 g f 4 g f g f gg:

5 Evnt Collction is a collction of all vnt occurrncs of a particular typ within a spcid tim intrval. It is dnotd by function } as follows. }(E start tim nd tim) =f jfg E[H] and start tim t occ() nd tim g: Givn a global vnt history, thvnt history for an arbitrary composit vnt formulatd using th oprators dnd in sction. can b computd. Blow, w dn ths computations formally. This formulation will comput all occurrncs of a composit vnt (along with participating constitunt vnt occurrncs) for a nit H. This U is trmd th unrstrictd contxt. Th oprators, r, 4 ar all lft associativ.. Th Unrstrictd Contxt. (E re )[H]=f j E [H] [ E [H]g.. (E 4E )[H]=ff i j gjf i j g U U ((E [H] E [H]) [ (E [H] E [H])) and t occ( i ) t occ( j )g.. ANY(m E E E n )[H] =ff i j k gj t occ( i ) < t occ( j ) < < t occ( k ) and jf i j k gj = m n and f i j k g P (lmnt) g whr P is th powr st and lmnt is a mmbr of th st: E [H] U E [H] U U E m [H] ach i (prmutation) can b any i from to n with th rstriction that ach U E participating in th mrgd-cartsian product ( ) is distinct. ANY(m E )[H]=ff i j k gj t occ( i ) < t occ( j ) < < t occ( k ) jf i j k gj = m n and f i j k gp(e[h])g. and 4. (E E )[H] =ff i j gjt occ( i ) < t occ( j ) and f i j ge [H] U E [H]g. 5. A(E E E )[H] =ff i j gjt occ( i ) < t occ( j ) and f i j k ge [H] U E [H] U E [H]g. 6. P(E TI E )[H] =ff i tgj U for all f i k ge [H] E [H] and t occ( i ) < t occ( k ) t =tocc( i )+j TI for intgr j and t t occ( k )g. 7. :(E )[E E ][H]=ff U i k gj f i k ge [H] E [H] and }(E t occ( i ) t occ( k )) = g. Th dnition of th cumulativ oprators includ th accumulation of vnt occurrncs ovr an intrval. This rquirs th function } to collct th appropriat occurrncs. A* and P* ar dnd blow using}. 8. A (E E E )[H]= ff i }(E t occ( i ) t occ( k )) k gj i E [H] and k E [H] and t occ( i ) < t occ( k )g. 9. P (E TI E )[H]=ff i U k gj for all f i k ge [H] E [H] and t occ( i ) < t occ( k ) = ft j t =tocc( i )+j TI for intgr j and t t occ( k )gg. Blow, w illustrat th computation of a composit vnt X on a global history H according to th abov dnitions of oprators in th unrstrictd contxt. Th vnt X is drawn from th stock markt applications. Th intrprtation of constitunt vnts of X is, E : opning of stock markt, E :chang in Dow Jons avrag, E : chang in th pric of IBM stock, and E 4 : chang in a commodity which dpnds on IBM stock. X =((E 4E ) E (E 4E 4 )) H = ff g f g f g f g f g f 4g f g f 4gg E [H] =ff g f gg E [H] =ff g f gg E [H] =ff g f gg E 4 [H] =ff 4g f 4gg (E 4E )[H]=ff g f g f g f gg (E 4E 4 )[H]=ff g f g f g f gg X[H] =ff g f g f g f g f g f g f g f g f g f g f g f g f g f g f g f gg As can b visualizd, thr ar 6 occurrncs of th vnt X for th givn history. It is not clar whthr all ths occurrncs will b usful in all applications. W strongly bliv that an application would b intrstd in a subst of ths vnts that ar maningful to

6 th smantics of that application. Furthrmor, dirnt applications may b intrstd in dirnt substs. In th nxt sction, w propos paramtr contxts as a way of imposing maningful rstrictions of th composit vnt history gnratd for an vnt. Not that th dtction a composit vnt in th unrstrictd contxt may warrant kping all vnt occurrncs (spcially for, Any, and 4 oprators) and hnc poss practical problms for th managmnt of vnt history and dtction. In most applications, ithr th tim intrval within which th vnts nd to b dtctd or th rlvanc of multipl occurrncs of th sam vnt is drivd from th application smantics. Hnc, only a subst of th vnts dtctd in th unrstrictd contxt is likly to b maningful. 4 Composit Evnt Dtction Evnts can always b dtctd and paramtrs computd using th unrstrictd contxt prsntd in th prvious sction. Howvr, th unrstrictd contxt producs a larg numbr of vnt occurrncs and not all occurrncs may b maningful from th point of viw of an application. Morovr, th computation and spac ovrhad associatd with th dtction of vnts in this contxt can b substantial. In this sction, w rn paramtr contxts introducd in [CM94] for th purpos of rducing th spac and computation ovrhad associatd with th dtction of composit vnts and providing a mchanism for choosing a maningful subst of vnt occurrncs gnratd by th unrstrictd contxt. Paramtr contxts srv th purpos of dtcting and computing th paramtrs of composit vnts in dirnt ways to match th smantics of applications. Th choic of a paramtr contxt also suggsts th complxity of vnt dtction and storag rquirmnts for a givn application. Th dtction of a composit vnt may rquir th dtction of on or mor constitunt vnts as wll as on or mor occurrncs of a constitunt vnt. Evnts rquiring multipl vnt occurrncs (ithr of th sam typ or of dirnt typs) for th dtction of a composit vnt, giv ris to altrnat ways of computing th history as wll as paramtrs, as th vnts ar likly to occur svral tims ovr an intrval. Th occurrnc of a composit vnt is markd by th occurrnc of a constitunt vnt that maks th composit vnt occur (using th nd-of vnt modir smantics). This constitunt vnt is trmd th trminator of th composit vnt. Svral constitunt vnts can act as trminators, but thr is at last on trminator vnt for a givn composit vnt. Analogously, thr is always a constitunt vnt that initiats th occurrnc of a composit vnt. This constitunt vnt is trmd th initiator of th composit vnt. Thr may b mor than on initiators for a composit vnt. For a primitiv vnt, th primitiv vnt itslf is th trminator and initiator. Th composit vnt dtctor nds to rcord th oc- E A ; E E tim 4 4 Figur : Global vnt history currncs of ach constitunt vnt andsav its paramtrs so that thy can b usd to comput th paramtr st of th composit vnt. Considr th followingvnt xprssions: A =(E 4E ) E B = E re re C = E ANY( E E ) whr E, E,andE ar primitiv vnts. Evnt A is dtctd whn E occurs providd both E and E hav alrady occurrd in any ordr. Evnt B is signald ach tim an instanc of any ofththrvnts E, E or E occurs. Paramtrs of vnt A (as wll as C) includ paramtrs of all th thr vnts E, E and E whras th paramtrs of vnt B includ only th paramtrs of on of its vnts. Both E and E can b initiators of A and E is th only trminator. For C, E is th initiator and both E and E can b trminators. Figur shows a global vnt history in which four typs of vnts E, E, E, and E 4 occur, as wll as th vnt graph for th composit vnt A. 4. Paramtr Contxts Th paramtr contxts proposd blow ar motivatd by a carful analysis of svral classs of applications. Whavidntid four paramtr contxts that ar usful for a wid rang of applications. Blow, w indicat th charactristics of th applications that motivatd our choic of paramtr contxts:. Applications whr th vnts ar happning at a fast rat and multipl occurrncs of th sam typ of vnt only rn th prvious data valu. In othr words, th ct of th occurrnc of svral vnts of th sam typissubsumdby th most rcnt occurrnc. This is typical of snsor applications (.g., hospital monitoring, global position tracking, multipl rmindrs for taking an action).. Applications whr thr is a corrspondnc btwn dirnt typs of vnts and thir occurrncs and this corrspondnc nds to b maintaind. Applications that xhibit causal dpndncy (.g., btwn aborts, rollbacks, and othr oprations btwn bug rports and rlass start of a transaction and its nd) com undr this catgory.. Trnd analysis and forcasting applications (.g., scuritis trading, stock markt, aftr-th-fact diagnosis) whr composit vnt dtction along a moving tim window nds to b supportd. For xampl, computing chang of mor than 0% in DowJons

7 avrag in any hour priod rquirs ach chang to initiat a nw occurrnc of an vnt. This corrsponds to th initiation of th dtction of an vnt for ach distinct occurrnc. 4. Applications whr multipl occurrncs of a constitunt vnt nds to b groupd and usd in a maningful way whn th vnt occurs. This contxt is usful in applications whr an vnt is trminatd by a dadlin-vnt and all occurrncs of constitunt vnts ar maningful up to th occurrnc of th dadlin vnt. For xampl, in a banking application w might want to kp track of th amount of withdrawals and dposits prformd in a day and us it to updat th account balanc at th nd of th day. W introduc th following contxts for th classs of applications dscribd abov. Ths contxts ar prcisly dnd using th notion of initiator and trminator vnts. W xplain th contxts using th composit vnt A which is a constitunt vnt of th composit vnt X in th prvious sction. W ar not concrnd with occurrncs 4 and 4 as vnt E 4 is not part of th vnt xprssion of A. Rcnt: In this contxt, only th most rcnt occurrnc of th initiator for any vnt that has startd th dtction of that vnt isusd. Whnanvnt occurs, th vnt is dtctd and all th occurrncs of vnts that cannot b th initiators of that vnt in th futur ar dltd (or ushd). For xampl, in th rcnt contxt, paramtrs of vnt A will includ th vnt instancs f,, g (A is dtctd whn occurs) and f,, g (whn A is dtctd again whn occurs). In this contxt, not all occurrncs of a constitunt vnt will b usd in dtcting a composit vnt. Furthrmor, an initiator of an vnt (primitiv or composit) will continu to initiat nw vnt occurrncs until a nw initiator occurs. Chronicl: In this contxt, for an vnt occurrnc, th initiator, trminator pair is uniqu. Th oldst initiator is paird with th oldst trminator for ach vnt (i.., in chronological ordr of occurrnc). Whn a composit vnt is dtctd, its paramtrs ar computd by using th oldst occurrnc of ach constitunt vnt. Howvr, onc usd occurrncs of th constitunt vnts cannot participat in any othr occurrncs of th composit vnt. For xampl, paramtrs of vnt A in th chronicl contxt will b computd by using vnt instancs f, and g. Whn th nxt E typ vnt occurs at, thn th A will b dtctd with th instancs f g. Continuous: In this contxt, ach initiator of an vnt starts th dtction of that vnt. A trminator vnt occurrnc may dtct on or mor occurrncs of th sam vnt. This contxt is spcially usful for tracking trnds of intrst on a sliding tim point govrnd by th initiator vnt. In Figur, ach of th occurrncs and (as wll as and )would start th dtction of th vnt A. Th rst occurrnc of A will hav th instancs f,, g. Th scond occurrnc of A will consist of f,, g. In this contxt, an initiator will b usd at last onc for dtcting that vnt. Thr is a subtl dirnc btwn th chronicl and th continuous contxts. In th formr, pairing of th initiator is with a uniqu trminator of th vnt whras in th lattr multipl initiators ar paird with a singl trminator of that vnt. Cumulativ: In this contxt, for ach constitunt vnt, all occurrncs of th vnt ar accumulatd until th composit vnt isdtctd. Whnvr a composit vnt is dtctd, all th constitunt vnts occurrncs that ar usd for dtcting that composit vnt ar dltd. For xampl, paramtrs of vnt A will includ all th instancs of ach vnt upto whn it occurs. Th ntir instancs shown in Figur (xcpt,and ) is th st of occurrncs that mak th composit vnt A. Unlik th continuous contxt, an vnt occurrnc dos not participat in two distinct occurrncs of th sam vnt in th cumulativ contxt. Obsrv that th cumulativ contxt dscribd abov cannot b gnratd as a subst of th vnt history gnratd by th unrstrictd contxt. Th notion of accumulation of vnt occurrncs is not prsnt in th unrstrictd contxt. For this rason, th dnitions of A* and P* usd th function } which accumulats a st of vnt occurrncs of a spcic typ ovr a givn intrval. Although contxts dscribd abov rstrict th st of vnt occurrncs gnratd, thy ar basd on th us of initiator, trminator pair in dirnt ways. In addition to th abov contxts, it may b usful to dtct composit vnts ovr non-ovrlapping tim intrvals. That is for any two occurrncs of an vnt W,th t occ of th initiator is gratr than th t occofthtrminator of th immdiatly prcding occurrnc of W. This notion of th us of non-ovrlapping intrvals can b applid to any of th contxts dscribd in this papr, including th unrstrictd contxt. This can b asily sn from th Figur. For instanc, all vnts dtctd in rcnt, chronicl, and continuous contxts ar not disjoint. If disjoint dtction of vnt occurrncs wr to b spcid for th xampl shown in Figur, only th rst occurrncs of vnts in ach contxt (i..,,, 5, and 9) would b dtctd. Basd on th abov dnitions of contxts, svral obsrvations can b mad. Disjoint continuous contxt is th sam as disjoint chronicl contxt. Also, cumulativ contxt always gnrats occurrncs that satisfy th disjoint spcication. In othr words, disjoint cumulativ contxt is quivalnt to cumulativ contxt.

8 Rcnt.. Chronicl. 4. Continuous Cumulativ 9. tim - Tim intrval ovr which X is dtctd - Initiator - Trminator - Participating primitiv vnts in th dtction of X Figur : Illustration of vnt dtction in various contxts for th xprssion X =(E 4E E E 4E 4 ) 4. Illustration of Composit Evnt Dtction Th approach takn for composit vnt dtction in this papr is dirnt from th approachs takn in Od and Samos. Samos dns a mchanism basd on Ptri Nts for modling and dtction of composit vnts for an OODBMS. Thy us modid colord Ptri nts calld SAMOS Ptri Nts to allow owof information about th vnt paramtrs in addition to occurrnc of an vnt. It appars that common subxprssions ar rprsntd sparatly lading to duplication of Ptri Nts. Furthrmor, although not statd xplicitly, Samos dtcts vnts only in th chronicl contxt dscribd in this papr. Od uss an xtndd nit automata for composit vnt dtction. Thir xtndd automaton, maks a transition at th occurrnc of ach vnt in th history lik a rgular automaton and in addition to that it looks at th attributs of th vnts, and also computs a st of rlations at th transition. Th dnitions of And and Pip oprators on vnt historis do not sm to produc th dsird rsult. W us an vnt tr for ach composit vnt and ths trs ar mrgd to form an vnt graph for dtcting a st of composit vnts. This will avoid th dtction of common sub-vnts multipl tims thrby rducing storag rquirmnts. Primitiv vnt occurrncs ar injctd at th lavs and ow upwards analogous to a data-ow computation. Furthrmor, th commonality of rprsntation btwn vnt dtction and qury optimization using oprator trs allow usto combin both, and optimiz a situation (vnt-condition pair) as a unit. This is crtainly possibl in th rlational modl as transformations can b applid to push prdicats from conditions to and apply thm during vnt dtction as part of th optimization (in contrast, vnt masks ar spcid in Od by th usr). Finally, th combination of vnt-condition trs will allow conditions to b valuatd on a dmand basis avoiding unncssary computations. In summary, our formulation of vnt dtction radily lnds itslf to optimization tchniqus usd in databass. Th introduction of paramtr contxts adds anothr prspctiv to th dtction of composit vnts. From Figur it is asir to undrstand how ach paramtr contxt dtcts dirnt instancs of th sam composit vnt foragivn squnc of primitiv vnt occurrncs. In this sction w will us on vnt graph and discuss how w comput th constitunt vnts of a composit vnt for ach of th paramtr contxts. Algorithms for dtcting composit vnts in dirnt contxts and thir implmntation ar dtaild in [Kri94]. Th tim lin indicats th rlativ ordr of th primitiv vnts with rspct to thir tim of occurrncs. All vnt propagations ar don in a bottom-up fashion. Th lavs of th graphs hav no storag and hnc pass th primitiv vnts dirctly to thir parnt nods. Th oprator nods hav sparat storag for ach of thir childrn. Th graphs shown in Figur for th various contxts ar at a tim point whn primitiv vnt 4 is dtctd. Th dirnt instancs of th sam vnt ar stord as sparat ntris and ar shown in sparat lins in th gur. Sinc th lavs do not hav any storag, th primitiv vnt 4 is passd to th parnt of laf E 4. Th arrows pointing from th child nod to its parnt in th graph indicats th dtction and ow of th vnts. In th rcnt contxt f g is snt tonodasinc and 4 ar th most rcnt initiator and trminator of th AND oprator (nod C). Sinc th trminator 4 can srv as an initiator for nod C (according to th smantics of AND), it is not discardd. At nodath initiator is alrady prsnt and f g srvs as th trminator. So vnt X is dtctd with f g. Hr sinc th trminator cannot srv as th initiator it is discardd and only f g which is th most rcnt initiator of X is rtaind at nod A. In th cas of Chronicl contxt, is th oldst initiator of nod C and it is at th had of th initiator list. Hnc 4 is paird with and f g is passd to nod A. Onc thy ar passd, unlik th rcnt contxt, both th initiator and th trminator ar discardd. Hnc nod C rtains only aftr AND is dtctd. Evnt X is dtctd with f g at nod A and both f g and f g ar dltd. Continuous contxt involvs lot of storag ovrhad for vnt dtction. As in th chronicl contxt w rtain all th initiators signalld so far in ach ofth nods. But unlik chronicl contxt, th trminator is paird with ach of th initiators prsnt and all th

9 ^ D ^ E E E4 E E B ; D ^ B ; E dtct X A ; ^ C Rcnt contxt E dtct X A 4 ; C E E E4 Continuous contxt 4 4 E D ^ B ; E E 4 ^ 4 E E E4 Chronicl contxt dtct X B C ; ^ 4 D ^ E E E4 E ; tim dtct X 4 A C A ; Cumulativ contxt Figur : Evnt dtction in various contxts initiators ar dltd aftr th dtction of th composit vnt. W rtain th trminator only if it can srv as an initiator for futur dtction of th composit vnt. At any point of tim, th trminator of th composit vnt X in all th othr contxts will signal only on occurrnc of vnt X, whras in th continuous contxt it will gnrat multipl occurrncs of X. In our xampl, ar th initiators at nod C. Both of thm ar paird with 4 to gnrat two occurrncs of th AND at th sam point of tim, namly f g, f g. Sinc 4 can srv as an initiator for nod C in th dtction of a nw occurrnc of th constitunt vnt, w rtain it and both th initiators, that hav bn paird ar dltd. At nod A, thr ar two initiators alrady prsnt andthtwo trminators signalld from nod C lad to four instancs of th dtction of vnt X with th sam tim of occurrnc. Among th four contxts prsntd, th continuous contxt gnrats a largr subst of th vnt occurrncs idntid by th unrstrictd cas. In th cumulativ contxt, unlik th continuous contxt, all th initiator occurrncs availabl so far ar combind with th trminator and only on occurrnc of X is dtctd. In our xampl,, ar combind togthr as on initiator and f g is snt to parnt nod A. Similarly, nod A dtcts X with f g. Onc dtctd th unid initiator and trminator is discardd. 4. Storag Rquirmnts Paramtr contxts dscribd in this papr simplify th vnt dtction as wll as th computation of paramtrs as compard to th unrstrictd contxt. Som of th paramtr contxts, such ascontinuous and chronicl, impos mor storag rquirmnts than th rcnt and cumulativcontxts. Th rcnt paramtr contxt can b implmntd using a xd siz bur for ach vnt (i.., at ach nodofthvnt graph). This is bcaus only th paramtrs for th most rcnt occurrnc of an vnt is stord and hnc rquirs th last amount of storag. For th chronicl contxt, a quu is rquird and th amount of storag ndd is dpndnt upon th duration of th intrval of th composit vnt and th frquncy of vnt occurrncs within that intrval. Similarly, for th continuous contxt, th storag rquirmnts can b xcssiv, implying that th choic of th paramtr contxt for ach rul nds to b mad judiciously. Th cumulativ contxt, unlik th continuous and chronicl contxts, combins all initiators and hnc at ach nod thr is only on whol initiator combination. Though both continuous and chronicl maintain a list of initiators, only continuous can signal mor than on occurrnc of a composit vnt for a singl trminator. Sinc this composit vnt might b a constitunt vnt of anothr largr xprssion, th continuous paramtr contxt rquirs considrabl storag compard to any othr paramtr contxt. Th storag rquirmnts can b xcssiv for th cumulativ contxt also. Howvr, basd on th smantics of th paramtr contxts, th storag rquirmnt incrass monotonically from rcnt tocumulativ to chronicl to continuous to unrstrictd. This is bcaus all th vnt occurrncs usd in th dtction of a composit vnt ar dltd whn th vnt isdtctd in th cumulativ contxt whras in th chronicl contxt, initiator and trminator vnt occurrncs ar paird in th ordr of occurrncs and hnc mor vnts ar stord for longr duration. Application of th disjoint modir,onanycontxt (xcpt th cumulativ), furthr rducs th storag rquirmnts by allowing vnts to b discardd arlir. 5 Activ OODBMS Architctur It is usful to xamin th rquirmnts of rul procssing in activ databass bfor prsnting an architctur. Broadly, th rquirmnts ar: 5. Support for Evnts Primitiv and Composit vnt dtction: Any mthod of any objct class is a potntial primitiv vnt. Furthr w prmit bfor- and aftr-variants of mthod invocation as vnts. Composit vnts ar formd by applying a st of oprators to primitiv vnts and composit vnts. Both primitiv and composit vnts nd to b dtctd by th

10 systm. Th dtction of composit vnts ntails not only th tim at which th composit vnt occurs, but also kping track of th constitunt vnt occurrncs. Paramtr computation: Th paramtrs of a primitiv vnt corrsponds to th paramtrs of th mthod dclard as a primitiv vnt. Th procssing of composit vnts ntails not only its dtction, but also th computation of th paramtrs associatd with a composit vnt. Th paramtrs of a composit vnt nd to b collctd, rcordd and passd on to condition and action portions of a rul by th vnt dtctor. Furthrmor, ths paramtrs nd to b rcordd in such a way that thy can b intrprtd by th condition and action componnts of a rul. Onlin and batch dtction of composit vnts: Th composit vnt dtctor nds to support dtction of vnts as thy happn (onlin) whn it is coupld to an application or ovr a stord vnt-log (in batch mod). Intr-application (global) vnts: In addition to ruls basd on vnts from within an application, it is usful to allow composit vnts whos constitunt vnts com from dirnt applications. This is spcially usful for cooprativ transactions and workow applications. This ntails dtction of vnts that span svral applications. 5. Support for Ruls Multipl ruls: An vnt (primitiv as wll as composit) can triggr svral ruls. Hnc, it is ncssary to support a rul xcution modl that supports concurrnt as wll as prioritizd rul xcution. Nstd ruls: Whn rul actions rais vnts which triggr othr ruls thr is nstd xcution of ruls. Ruls can b nstd to arbitrary lvls. Coupling mods: Th thr coupling mods (immdiat, dfrrd and dtachd) discussd in HiPAC wr introducd to support application nds. Sntinl architctur should b abl to support all of thm. Rul schduling: In th prsnc of multipl ruls and nstd xcution, th architctur nd to support prioritizd srial xcution of ruls, concurrnt xcution of all ruls, or a combination of th two. Furthr, th systm, should allow th application dsignr to choos from among th abov altrnativs. Th abov rquirmnts as wll as th OO modl into which activ capability is bing incorporatd act th dsign of both th rul procssing subsystm and th vnt dtctor. Blow, w prsnt th Sntinl architctur in trms of xtnsions to th Opn OODB systm and discuss how th abov rquirmnts ar supportd in our currnt implmntation. 5. Sntinl Architctur Th Sntinl architctur proposd in this sction xtnds th passiv Opn OODB systm [Ins9]. In ordr to satisfy th abov rquirmnts in an objct-orintd framwork, w propos th architctur shown in Figur which is bing implmntd as an xtnsion to th Opn OODB Toolkit dvlopd at Txas Instrumnts. Our proposd architctur rlis on th us of thrads (or light wight procsss) for sparating vnt dtction from application xcution in a transparnt mannr. Our primitiv vnt dtction is basd on th dsign proposd in [AMC9]. Primitiv vnts ar signald by adding a notify procdur call in th wrappr mthod by Sntinl. Also, appropriat calls for th paramtr collction ar addd at this stag. Both primitiv and local composit vnts ar signald as soon as thy ar dtctd. Howvr, th dtction of a composit vnt may span a tim intrval as it involvs th dtction and grouping of its constitunt vnts in accordanc with th paramtr contxt spcid. A clan sparation of th dtction of primitiv vnts (as an intgral part of th databas) from that of composit vnts allows on to: i) implmnt a composit vnt dtctor as a sparat modul (as has bn don) and ii) introduc additional vnt oprators without having to modify th dtction of primitiv vnts. Each application has a local composit vnt dtctor (Figur) to which all primitiv vnts ar signald. Our implmntation uss thrads (light wight procsss), instad of procsss, for sparating composit vnt dtction (as wll as for th xcution of ruls) from application. Whn a primitiv vnt occurs it is snt to th local composit vnt dtctor and th application waits for th signaling of a composit vnt that is dtctd in th immdiat mod. Th local composit vnt dtctor and th application shar th sam addrss spac and our vnt dtctor uss an vnt graph similar to oprator trs. Paramtr computation for composit vnts raiss additional problms in th objct-orintd framwork. Th lack of a singl data structur (such as a rlation) maks it xtrmly dicult to idntify and manag paramtr computation vn within an application. As a rst cut, w includ th idntication of th objct (i.., oid) as on of th vnt paramtrs and othr paramtrs which hav atomic valus. Howvr, no assumptions ar mad about th stat of th objct (whn th oid is passd as part of a composit vnt) as th dtction of a composit vnt isovr a tim intrval. A linkd list that contains th paramtrs of ach primitiv vnt (as a list) that participats in th dtction of th composit vnt is computd and passd to th rul associatd with that vnt. Complt support for paramtrs of composit vnts may rquir vrsioning of objcts and rlatd concurrncy control and rcovry tchniqus. A rul spcid to b xcutd in th dfrrd mod

11 Global Evnt Dtctor Global Evnts Application to xcut dtachd rul Forkd procss Application n to xcut dtachd rul Forkd procss Global Evnts Local Evnt Dtctor bgin bgin Transaction Transaction H H E a E a v n v n 6 d d 6 n l n l t t r r nd Transaction Application nd Transaction Application N Local Evnt Dtctor - Primitiv Evnt signald - pr-commit and abort signald 5 - Intr-application vnts dtctd - Composit vnt dtction for immdiat ruls 4 - Causally dpndnt commit signald 6 - Ruls xcutd as subtransactions Figur: Sntinl architctur is rwrittn at th sourc cod lvl into a rul in immdiat mod by th Sntinl pr-procssor. Our vnt spcication languag Snoop [CM94] supports a numbr of oprators of which A monitors th cumulativ ct of an vnt occurrnc within a spcid intrval. For xampl, if w nd to accumulat all insrt vnts in a transaction, w can spcify th vnt as A (bgin transaction, insrt, nd transaction). Using this oprator, w modl th dfrrd coupling mod in trms of th immdiat coupling mod by using bgin and pr-commit transaction vnts and postpon th xcution of th rul to th nd of th transaction. In Sntinl, th bgin transaction vnt isalways signalld at th bginning of a transaction and th prcommit is signalld bfor th commit of a transaction. Using th A oprator, a rul in dfrrd mod with an (arbitrary) vnt E is transformd by th Sntinl pr-procssor to A (bgin-transaction, E, pr-committransaction). This causs a dfrrd rul to b xcutd xactly onc vn though its vnt may b triggrd a numbr of tims in th cours of that transaction xcution. For rul xcution, a nstd transaction managr is implmntd with its own lock managr. This is in addition to th concurrncy control and rcovry providd by th Exodus for top-lvl transactions. Each rul (i.., condition and action portions of a rul) is packagd into a subtransaction. A numbr of subtransactions ar spawnd as a part of th application procss. This is furthr laboratd in [CKTB94]. Support for multipl rul xcution and nstd rul xcution ntails that th vnt dtctor b abl to rciv vnts dtctd within a rul's xcution in th sam mannr it rcivs vnts dtctd in a top lvl transaction. This is accomplishd rlativly asily by sparating th local composit vnt dtction from th application as shown in Figur. This sparation also radily supports both onlin and batch (or aftr-th-fact) dtction of composit vnts. Finally, in th prsnc of composit vnts, it is possibl for th vnts to cross transaction boundaris (within th sam application). Currntly, w provid a mchanism to ush all vnts gnratd by a transaction whn it commits. Mor work is rquird to undrstand th smantics of rul xcution whos vnts span transaction boundaris. 6 Conclusions This papr signicantly xtnds our arlir work on an xprssiv vnt spcication languag. W hav providd a dclarativ smantics of ach oprator of Snoop. W introducd th notion of global vnt history and local vnt history for dning th computation of participating vnts for an arbitrary composit vnt xprssion. W rnd th paramtr contxts introducd arlir using initiator and trminator vnts. W hav illustratd th dtction of composit vnts in various

12 contxts and proposd an architctur for its implmntation in an objct-orintd framwork along with a discussion of th various issus involvd. Finally, algorithms for all paramtr contxts hav bn dvlopd and implmntd using th architctur shown in this papr. In this papr, w ar assuming that th paramtrs of an vnt can b computd onc th vnt occurrncs ar known. It is usful, howvr, to xplicitly introduc (as a minimum) th idntication of th objct (i.., oid) for which th primitiv vnt is applicabl. This can b don by spcifying, for ach primitiv vnt, a paramtr which is ithr a constant oravariabl rprsnting th oid. For xampl, th primitiv vnt Chang pric(ibm) indicats that th vnt occurs whn th mthod Chang pric is xcutd for th IBM objct. As anothr xampl, Chang pric(x) Chang pric(x) rfrs to th squnc of vnts on th sam oid X. And Chang pric(x) Chang pric(y) rfrs to th squnc of vnts on two dirnt oid's. Dtaild discussion of paramtr computations ar byond th scop of this papr. Som of ths issus hav bn discussd in sction 5. All th vnt dtction algorithms w hav dvlopd xtnd radily whn th oid is allowd as an xplicit paramtr of a primitiv vnt. Acknowldgmnts This work is supportd, in part, by th National Scinc Foundation IRI-906, by th Oc of Naval Rsarch and th Navy Command, Control and Ocan Survillanc Cntr RDT&E Division, and by throm Laboratory. Rfrncs [AMC9] E. Anwar, L. Maugis, and S. Chakravarthy. A Nw Prspctiv on Rul Support for Objct-Orintd Databass. In Procdings, Intrnational Confrnc on Managmnt of Data, pags 99{08, Washington, D.C., May 99. [CKTB94] S. Chakravarthy, V. Krishnaprasad, Z. Tamizuddin, and R. Badani. ECA Rul Intgration into an OODBMS: Architctur and Implmntation. Tchnical Rport UF-CIS-TR-9-09, Univrsity of Florida, E470-CSE, Gainsvill, FL 6, Fb (Submittd for publication.). [CM94] S. Chakravarthy and D. Mishra. Snoop: An Exprssiv Evnt Spcication Languag for Activ Databass. Data and Knowldg Enginring, 994. (To appar). [FM87] C. L. Forgy and J. McDrmott. Domain- Indpndnt Production Systm Languag. In Procdings Fifth Intrnational Confrnc on Articial Intllignc, Cambridg, MA, 987. [For8] C. L. Forgy. RETE: A Fast Algorithm for th Many Pattrn/Many Objct Pattrn Matching Problm. Articial Intllignc 9, pags 7{7, 98. [GD9] S. Gatziu and K. R. Dittrich. Evnts in an Objct-Orintd Databas Systm. In Proc. of th st Intrnational Confrnc onruls in Databas Systms, Sptmbr 99. [GD94] S. Gatziu and K. R. Dittrich. Dtcting Composit Evnts in Activ Databass Using Ptri Nts. In Proc. of thth Intrnational Workshop on Rsarch Issus in data Enginring: Activ Databas Systms, pags {9, Fbruary 994. [GJS9a] N. H. Ghani, H. V. Jagadish, and O. Shmuli. COMPOSE A Systm For Composit Evnt Spcication and Dtction. Tchnical rport, AT&T Bll Laboratoris, Murray Hill, NJ, Dcmbr 99. [GJS9b] N. H. Ghani, H. V. Jagadish, and O. Shmuli. Evnt Spcication in an Objct-Orintd Databas. In Procdings, Intrnational Confrnc on Managmnt of Data, pags 8{90, San Digo, CA, Jun 99. [Ins9] Txas Instrumnts. Opn OODB Toolkit, Rlas 0. (Alpha) Documnt, Sptmbr 99. [Kri94] V. Krishnaprasad. Evnt Dtction for Supporting Activ Capability in an OODBMS: Smantics, Architctur, and Implmntation. Mastr's thsis, Databas Systms R&D Cntr, CIS Dpartmnt, Univrsity of Florida, E470-CSE, Gainsvill, FL 6, March 994. [Mis9] D. Mishra. SNOOP: An Evnt Spcication Languag for Activ Databass. Mastr's thsis, Databas Systms R&D Cntr, CIS Dpartmnt, Univrsity of Florida, E470-CSE, Gainsvill, FL 6, August 99. [SW9] A. Prasad Sistla and O. Wolfson. Tmporal Triggrs in Activ Databass. Tchnical rport, Univrsity of Illinois at Chicago, Chicago, IL, July 99.