"... In PAGE 30: ... All system characteristics which are contained in the classification framework are explicitly chosen by the system or protocol designer. Table3 -1 classifies these protocols. A different classification for replica update protocols, which concentrates on relationships between the protocols, is presented in [CHKS94].... In PAGE 34: ...Chapter 3: State of the art Table3 -1: Classification of replication protocols. Replication transparency Transparent Non-trans- parent Not applicable 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, QCP, PT, ES, BI, Esc, VC Consistency Strong 2PC, 3PC, QC, ROWA, MW, VP, Re, VC1 Weak AE, TSAE, LR, IUIA, DP, MC, TU, QCP, PT, ES, BI, Esc, VC2 Replica syn- chronization Synchronous 2PC, 3PC, QC, ROWA, MW, VP, MC3, Re5, VC1 As soon as possible AE, TSAE, LR, IUIA, MC4, BI, Esc Temporal event QCP10, PT, ES11, VC2 Non-tempo- ral event DP7, TU8, Re6, QCP10, ES11 Update rights Master-slave TU, QCP, VC8 Peer-to-peer 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, PT, ES, BI, Esc Conflicts Conflicts can occur AE, TSAE9, LR9, IUIA, DP, MC, TU, PT, ES11, BI12 No conflicts occur 2PC, 3PC, QC, ROWA, MW, VP, TSAE9, LR9, Re, QCP, Esc, VC, ES11, BI12 Implicit replica- tion of refer- enced objects Yes No Not applicable 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, QCP, PT, ES, BI, Esc, VC Implicit schema replication Yes Partial No Not applicable 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, QCP, PT, ES, BI, Esc, VC Dynamic allo- cation of repli- cas Yes No Not applicable 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, QCP, PT, ES, BI, Esc, VC Data model Object- oriented Rela- tional Files Other data model Not applicable or open 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, QCP, PT, ES, BI, Esc, VC 3.... In PAGE 34: ... Replication transparency Transparent Non-trans- parent Not applicable 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, QCP, PT, ES, BI, Esc, VC Consistency Strong 2PC, 3PC, QC, ROWA, MW, VP, Re, VC1 Weak AE, TSAE, LR, IUIA, DP, MC, TU, QCP, PT, ES, BI, Esc, VC2 Replica syn- chronization Synchronous 2PC, 3PC, QC, ROWA, MW, VP, MC3, Re5, VC1 As soon as possible AE, TSAE, LR, IUIA, MC4, BI, Esc Temporal event QCP10, PT, ES11, VC2 Non-tempo- ral event DP7, TU8, Re6, QCP10, ES11 Update rights Master-slave TU, QCP, VC8 Peer-to-peer 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, PT, ES, BI, Esc Conflicts Conflicts can occur AE, TSAE9, LR9, IUIA, DP, MC, TU, PT, ES11, BI12 No conflicts occur 2PC, 3PC, QC, ROWA, MW, VP, TSAE9, LR9, Re, QCP, Esc, VC, ES11, BI12 Implicit replica- tion of refer- enced objects Yes No Not applicable 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, QCP, PT, ES, BI, Esc, VC Implicit schema replication Yes Partial No Not applicable 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, QCP, PT, ES, BI, Esc, VC Dynamic allo- cation of repli- cas Yes No Not applicable 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, QCP, PT, ES, BI, Esc, VC Data model Object- oriented Rela- tional Files Other data model Not applicable or open 2PC, 3PC, QC, ROWA, MW, VP, AE, TSAE, LR, IUIA, DP, MC, TU, Re, QCP, PT, ES, BI, Esc, VC 3.1 Replication protocols 41 Table3 -2: Legend to table 3-1. 2PC Two-phase commit MW Missing writes 3PC Three-phase commit PT Polytransaction AE Anti-entropy QC Quorum consensus BI Bounded ignorance QCP Quasi-copy DP Data-patch Re Referee ES Epsilon-serializability ROW A Read-One-Write-All Esc Escrow TSAE Timestamped anti-entropy IUIA Independent updates and in- cremental agreement TU Tentative update LR Lazy Replication VC Virtual primary copy MC mc-compatibility VP Virtual partition 1 Within virtual primary copy 7 Initiated by database adminis- trator 2 Outside virtual primary copy 8 Update requests of weak repli- cas are redirected to the virtual primary copy 3 For commutative transactions 9 Depending on update ordering 4 For non-commutative transac- tions 10 Depending on coherency conditions 5 Within CSCR 11 Depending on replica control protocol 6 Outside CSCR 12 Depending on the global and local consistency constraints 3.... In PAGE 40: ...Chapter 3: State of the art Table3 -3: Classification of replication in distributed systems. Replication transparency Transparent Is, FW, RG, OT, IB, AFS, Co10, Fi, Fr, Hu, Ru, MI, Ar, El, Ro, Go, Ha, SOM, ARO, DV, GI, RMS Non-trans- parent Ca, Ba, Co11 Not ap- plicable Consistency Strong Is1,2, Fr3,4, Hu, Ar, El1,2, Ro9, Go, Ha, SOM, ARO12, DV19, RMS Weak Is1,2, FW, RG, OT, IB, AFS, Co, Fi, Fr4, Ru, MI, Ca, El1,2, Ro9, ARO12, DV19, GI, Ba Replica syn- chronization Synchronous Is1,2, Fr3,4, Hu, Ar, El1,2, Ro9, Go, Ha, SOM, ARO12, DV19, RMS As soon as pos- sible Is1,2, FW, RG, AFS, Co10, Fi, Fr4, El1,2, Ro9, ARO12, DV19, GI22 Temporal event Fr4, Ca, ARO12 Non-temporal event OT14, IB14, Co11, Fr4, Ru8, MI7, ARO12, GI23, Ba Update rights Master- slave Ca, SOM13 Peer-to-peer Is, FW, RG, OT, IB, AFS, Co, Fi, Fr, Hu, Ru, MI, Ar, El, Ro, Go, Ha, SOM13, ARO, DV, GI, RMS, Ba Conflicts Conflicts can occur Is1,2, FW1, RG1, OT16, IB18, AFS, Co, Fi, Fr4, Ru, El1,2, Ro9, GI, Ba No conflicts occur Is1,2, FW1, RG1, OT20, IB15, Fr3, Hu, MI, Ca, Ar, El1,2, Ro9, Go, Ha, SOM, ARO5, DV17, RMS Implicit replica- tion of refer- enced objects Yes DV21 No OT, IB, Ar, El, Ro, Go, Ha, SOM, ARO, GI, RMS Not applicable Is, Fw, RG, AFS, Co, Fi, Fr, Hu, Ru, MI, Ca, Ba Implicit schema replication Yes Partial OT6, IB6 No Ar, El, Ro, Go, Ha, SOM, ARO, DV, GI, RMS Not applicable Is, FW, RG, AFS, Co, Fi, Fr, Hu, Ru, MI, Ca, Ba Dynamic allo- cation of repli- cas Yes RMS No Is, FW, RG, OT, IB, AFS, Co, Fi, Fr, Hu, Ru, MI, Ca, Ar, El, Ro, Go, Ha, SOM, ARO, DV, GI, Ba Not appli- cable Data model Object-ori- ented OT, IB, Ar, El, Ro, Go, Ha, SOM, ARO, DV, GI, RMS Rela- tional Files AFS, Co, Fi, Fr, Hu, Ru, MI, Ca Other data model Not applicable or open Is, FW, RG, Ba 3.... In PAGE 40: ... Replication transparency Transparent Is, FW, RG, OT, IB, AFS, Co10, Fi, Fr, Hu, Ru, MI, Ar, El, Ro, Go, Ha, SOM, ARO, DV, GI, RMS Non-trans- parent Ca, Ba, Co11 Not ap- plicable Consistency Strong Is1,2, Fr3,4, Hu, Ar, El1,2, Ro9, Go, Ha, SOM, ARO12, DV19, RMS Weak Is1,2, FW, RG, OT, IB, AFS, Co, Fi, Fr4, Ru, MI, Ca, El1,2, Ro9, ARO12, DV19, GI, Ba Replica syn- chronization Synchronous Is1,2, Fr3,4, Hu, Ar, El1,2, Ro9, Go, Ha, SOM, ARO12, DV19, RMS As soon as pos- sible Is1,2, FW, RG, AFS, Co10, Fi, Fr4, El1,2, Ro9, ARO12, DV19, GI22 Temporal event Fr4, Ca, ARO12 Non-temporal event OT14, IB14, Co11, Fr4, Ru8, MI7, ARO12, GI23, Ba Update rights Master- slave Ca, SOM13 Peer-to-peer Is, FW, RG, OT, IB, AFS, Co, Fi, Fr, Hu, Ru, MI, Ar, El, Ro, Go, Ha, SOM13, ARO, DV, GI, RMS, Ba Conflicts Conflicts can occur Is1,2, FW1, RG1, OT16, IB18, AFS, Co, Fi, Fr4, Ru, El1,2, Ro9, GI, Ba No conflicts occur Is1,2, FW1, RG1, OT20, IB15, Fr3, Hu, MI, Ca, Ar, El1,2, Ro9, Go, Ha, SOM, ARO5, DV17, RMS Implicit replica- tion of refer- enced objects Yes DV21 No OT, IB, Ar, El, Ro, Go, Ha, SOM, ARO, GI, RMS Not applicable Is, Fw, RG, AFS, Co, Fi, Fr, Hu, Ru, MI, Ca, Ba Implicit schema replication Yes Partial OT6, IB6 No Ar, El, Ro, Go, Ha, SOM, ARO, DV, GI, RMS Not applicable Is, FW, RG, AFS, Co, Fi, Fr, Hu, Ru, MI, Ca, Ba Dynamic allo- cation of repli- cas Yes RMS No Is, FW, RG, OT, IB, AFS, Co, Fi, Fr, Hu, Ru, MI, Ca, Ar, El, Ro, Go, Ha, SOM, ARO, DV, GI, Ba Not appli- cable Data model Object-ori- ented OT, IB, Ar, El, Ro, Go, Ha, SOM, ARO, DV, GI, RMS Rela- tional Files AFS, Co, Fi, Fr, Hu, Ru, MI, Ca Other data model Not applicable or open Is, FW, RG, Ba 3.2 Replication in distributed systems 53 Table3 -4: Legend to table 3-3. AFS Andrew File System Go GOOFY Ar Arjuna Ha Hawk ARO Adaptable replicated objects Hu Huygens Ba Bayou IB Information Bus Ca Castanet Is Isis Co Coda MI MIo-NFS DV DistView OT OrbixTalk El ELECTRA RG Replication group Fi Ficus RMS Replica Management System Fr Frolic Ro ROMANCE FW Framework for group com- munication systems Ru Rumor GI GINA SOM SOM Replication Framework 1 Depending on update ordering 13 Depending on the number of writable replicas 2 Depending on reply collection 14 Pull- and push-actions are explicitly initiated 3 Between clusters 15 With guaranteed message de- livery 4 Within cluster: Depending on replica control strategy 16 Without persistent event chan- nels 5 If consistency manager disal- lows asynchronous peer-to- peer replication for conflicting updates 17 If lock object disallows asyn- chronous peer-to-peer replica- tion for conflicting updates 6 Implicit by self-describing format 18 Without guaranteed message delivery 7 On token passing 19 Depending on replica control strategy of the lock object 8 Reconciliation is explicitly initiated 20 With persistent event channels 9 Depending on replica control strategy 21 Only at replication setup time 10 In connected mode 22 With tight or loose coupling 11 In disconnected mode 23 In decoupled mode 12 Depends on consistency man-... In PAGE 44: ...Chapter 3: State of the art Table3 -5: Classification of replication in database systems. Replication transparency Transparent VR, DRO Non-transparent Gem, Sy, IBM, ER, OR, Ora, CA, SR, SQLS, Jet, IP, LN, OSC, OSAR Not appli- cable Consistency Strong VR, DRO Weak Gem, Sy, IBM, ER, OR, Ora, CA, SR, SQLS, Jet, IP, LN, OSC, OSAR Replica syn- chronization Synchronous VR, DRO As soon as possible Sy, ER, OR, Ora, CA, SQLS, OSC Temporal event Gem, IBM, ER, OR, Ora, CA, SQLS, Jet, IP, LN, OSAR Non-temporal event IBM, ER, OR, Ora, CA, SR, Jet, IP Update rights Master-slave Gem, Sy, IBM, ER, OR, Ora, CA, SQLS, Jet3, IP, OSAR Peer-to-peer VR, DRO, ER, OR, Ora, CA, SR, Jet4, LN, OSC Conflicts Conflicts can occur ER1, OR1, Ora1, CA1, SR, Jet4, LN, OSC No conflicts occur Gem, VR, DRO, Sy, IBM, ER2, OR2, Ora2, Ca2, SQLS, Jet3, IP, OSAR Implicit replica- tion of refer- enced objects Yes VRO, DRO, OSAR No Gem Not applicable Sy, IBM, ER, OR, Ora, CA, SR, SQLS, Jet, IP, LN, OSC Implicit schema replication Yes Jet, LN Partial No Gem, VR, DRO, Sy, IBM, ER, OR, Ora, CA, SR, SQLS, IP, OSC, OSAR Not appli- cable Dynamic allo- cation of repli- cas Yes No Gem, VR, DRO, Sy, IBM, ER, OR, Ora, CA, SR, SQLS, Jet, IP, LN, OSC, OSAR Not applicable Data model Object- oriented Gem, VR, DRO, OSAR Relational Sy, IBM, ER, OR, Ora, CA, SR, SQLS, Jet, IP Files IP Other data model IP, LN Not appli- cable or open OSC 3.... ..."

"... In PAGE 21: ....3. Future key technologies for manufacturing In the FutMan scenario exercise the experts identified key technologies for the industry sectors electronic components; measuring, testing, and control instruments; basic chemical industry and motor vehicles that are likely to shape manufacturing in the future. Table2 summarises the areas of new technology and their applications which were identified in the scenario exercise. Advances in science and technology, especially in materials science, microelectronics and information technology, biotechnology and nanotechnology will profoundly affect manufacturing and help manufacturers master the challenges ahead.... ..."

"... In PAGE 41: ...3m-19.1 million (Viscusi and Aldy, 2002, Table4 , pp92-93). Since there are relatively few Australian studies, there is also the issue of converting foreign (US) data to Australian dollars using either exchange rates or purchasing power parity and choosing a period.... ..."

"... In PAGE 42: ...1. Ideal Functional Form of Detector In Chapter 1 we introduced an ideal classifier as a large table as shown in Table5 below. This classifier is ideal in several ways.... In PAGE 45: ... To use such a representation, we must build a table that stores two values for each input, P(image | object) and P(image | non-object), as shown below in Table 7. There are several conse- quences of using a decision rule derived from this generalized form instead of one directly derived from the concise form in Table5 . The generalized form will work just as well if our probability estimates are accurate representations of the true probabilities.... In PAGE 51: ... This underscores the problem with naive Bayes classifiers or any classifier in general; that is, we will have errors when we do not model important relationships, such as that between x and y when P(x | y, object) and P(x | y, non-object) are very different functions. No classifier is immune to this problem except the ideal model, Table5 , which is infeasible. In a naive Bayes classifier, the relationships that are not modeled are very explicit.... ..."

"... In PAGE 9: ...Medical Prognosis Test Case The problem is to estimate the risk of death for patients in intensive care units, starting from a a set of physiological variables (see Table1 ) measured on a patient at its arrival in the hospital. The state-of-the-art method currently used for predicting the outcome, from this physiological variables, is based on the Simpli ed Acute Physiology Score (SAPS-II) [7], a statistical method for evaluating the severity of the patient status.... ..."

"... In PAGE 4: ............... 18 Table10 --Past and projected trends in use of cereal as feed, to the year 2020 19 Table 11--Total meat production, consumption, and net trade, 1993 and 2020 21 Table 12--Net exports (imports) of various meats by location in 1993 and projected to the year 2020 .... ..."

"... In PAGE 4: ..................................... 10 Table6 --Past and projected production trends of meat, to the year 2020 .... ..."

"... In PAGE 4: ................................................................ 13 Table8 --Past and projected production trends of various meats, to the year 2020 14 Table 9--Alternative projections of grain production in China .... In PAGE 15: ...han did the Unites States, a fact many in the U.S. may find surprising. Total world meat production is projected to increase 63 percent by 2020 over the levels of the early 1990s, which themselves were 31 percent over the levels of the early 1980s, as shown in Table8 . By 2020, China alone is projected to account for almost 30 percent of world meat production, at levels comparable to total production in the developing world (including China) in the early 1990s.... ..."

"... In PAGE 4: ........................................................ 6 Table4 --Past and projected consumption trends of meat, to the year 2020 .... In PAGE 12: ... By 2020, the developing countries will eat 70 percent more meat than the developed countries, although their per capita consumption is projected to be only 38 percent of developed country levels, compared to 20 percent in the early 1980s. The locus of this blockbuster change in world meat consumption patterns can be seen in Table4 . Meat consumption in China--accounting for over a fifth of the world apos;s population--grew at 8.... ..."