1、Introduction to Temporal Database Research,by Cyrus Shahabi from Christian S. Jensens Chapter 1,Outline,Introduction & definition Modeling Querying Database design Logical design Conceptual design DBMS implementation Query processing Implementation of algebraic operators Indexing structures Summary
2、Open problems,Introduction,Most applications of database technology are temporal in nature: Financial apps.: portfolio management, accounting & banking Record-keeping apps.: personnel, medical-record and inventory management Scheduling apps.: airline, car, hotel reservations and project management S
3、cientific apps.: weather monitoring,Definitions,Temporal DBMS manages time-referenced data, hence, times are associated with database entities Two types of time: valid time and transaction time Valid time, vt, of a fact (any logical statement that is either true or false) is the collected times (pos
4、sibly spanning the past, present & future) when the fact is true Although all facts have a valid time, the valid time of a fact may not necessarily be recorded in the database (unknown or irrelevant to the app.) If a database models different worlds, database facts might have several valid times, on
5、e for each world,Definitions ,Transaction time, tt: the time that a fact is current in the database Tt may be associated with any database entity, not only with facts Although all entities can be assigned a tt, the database designer may decide to not capture this aspect for some entities Tt aspect o
6、f an entity has a duration: from insertion to deletion, with multiple insertions and deletions being possible for the same entity Hence, deletion is pure logical (not physically removed but ceased to be part of the databases current state,Definitions ,Tt captures time varying states of the db & apps
7、. that demand accountability and tractability rely on dbs that record Tt Tt, unlike vt, is well-behaved and may be supplied automatically by the DBMS Both tt and vt values are drawn from a time domain, which may or may not stretch infinitely into the past and future Time domain may be discrete or co
8、ntinuous In databases, a finite and discrete time domain is typically assumed,Definitions ,Time is assumed to be totally ordered, but various partial orders and cyclic time has also been suggested Uniqueness of “Now”: the current time is ever-increasing, all activity is trapped at the current time,
9、and current time separates the past from the future The spatial equivalent “here” doesnt have the above properties; the biggest difference between time and space is that time cannot be reused! The uniqueness of now is one of the reasons why techniques from other research areas are not readily (or no
10、t at all) applicable to temporal data Now offers new data management challenges particular to temporal databases,Modeling,To extend a DBMS to become temporal, mechanisms must be provided for capturing valid and transaction times of the facts recorded by relations (temporal relations) More than 24 ex
11、tended relational models proposed to add time to relational model, most of which supported only valid time We consider three bitemporal ones for a video rental applications: customers check out tapes for certain durations of time and dates.,Modeling ,Bitemporal Conceptual Data Model (BCDM): timestam
12、ps tuples with sets of (tt, vt) values,C101 rents T1234 on May 2nd for 3 days, & returns it on 5th C102 rents T1245 on 5th open-ended, & returns it on 8th C102 rents T1234 on 9th to be returned on 12th. On 10th the rent is extended to include 13th but tape is not returned until 16th.,cID,TapeNum,C10
13、1,C102,C102,T1234,T1245,T1234,(2,2), (2,3), (2,4), (3,2), (3,3), (3,4), , (UC,2), (UC,3), (UC,4),(5,5), (6,5), (6,6), (7,5), (7,6), (7,7), (8,5), (8,6), (8,7), (UC,5), (UC,6), (UC,7),(9,9), (9,10), (9,11), (10,9), (10,10), (10,11), (10,12), (10,13), (13,9), (13,10), (13,11), (13,12), (13,13), (14,9)
14、, , (14,14), (15,9), , (15,15), (16,9), , (16,15), , (UC,9), , (UC,15),Modeling ,Bitemporal Conceptual Data Model (BCDM): timestamps tuples with sets of (tt, vt) values,C101 rents T1234 on May 2nd for 3 days, & returns it on 5th C102 rents T1245 on 5th open-ended, & returns it on 8th C102 rents T123
15、4 on 9th to be returned on 12th. On 10th the rent is extended to include 13th but tape is not returned until 16th.,1,17,5,1,5,9,9,Modeling ,BCDM pros: Since no two tuples with mutually identical explicit values are allowed in BCDM relation instance, the full history of a fact is contained in exactly
16、 one tuple Relation instances that are syntactically different have different information content and vice versa BCDM cons: Bad internal representation and display to users of temporal info Varying length and voluminous timestamps of tuples are impractical to manage directly Timestamp values are har
17、d to comprehend in BCDM format,Modeling ,Fixed-length format for tuples, where each tuples timestamp encodes a rectangular or stair-based bitemporal region Several tuples may be needed to represent a single fact,C101 rents T1234 on May 2nd for 3 days, & returns it on 5th C102 rents T1245 on 5th open
18、-ended, & returns it on 8th C102 rents T1234 on 9th to be returned on 12th. On 10th the rent is extended to include 13th but tape is not returned until 16th.,Modeling ,Non-first-normal-form representation, in which a relation is thought of as recording information about some types of objects (see pa
19、per) Note that 2nd tuple records two facts: rental information for customer C102 for the two tapes Pros of the two latter models: No need to update the relation at every tick, it is achieved by introducing “now” variable that assume the current value Two choices to enter time values into relations A
20、t the level of tuples (tuple timestamping) At the level of attribute values (attribute timestamping),Modeling ,Relation instances that all three models may record are snapshot equivalent (corresponding to a point-based view of data), e.g.,A,Vs,Ve,a,b,2,2,8,8,A,Vs,Ve,a,a,2,5,4,8,b,2,8,A,Vs,Ve,a,b,2,2
21、,8,4,b,5,8,The first relation is coalesced version of the other two, but they are snapshot equiv. Coalescing operation merges value equivalent tuples with same non-timestamp attributes and adjacent or overlapping time intervals,Modeling ,BCDM only allows coalesced relation instances, i.e., relations
22、 are only different if they are not snapshot equivalent The last two relations are not legal in BCDM However, the three relations are not equivalent from an interval-based view: First relation: a tape was checked out for 7 days Second relation: the tape was checked out for 3 days initially and then
23、for 4 more days,Querying,Temporal queries “can” be expressed via conventional query languages such as SQL (e.g., current temporal applications); however, with great difficulty,cID,TapeNum,C101,C102,C102,C103,T1234,T1425,T1324,T1243,cID,TapeNum,C101,C102,C102,C103,T1234,T1245,T1324,T1243,C101,T1245,C
24、102,T1425,C102,T1434,Vs,Ve,2,5,22,9,4,9,7,21,now,14,19,25,10,now,At time 17, the first relation is a snapshot of the second,S-CheckedOut,V-CheckedOut,Querying ,Number of current checkouts: SELECT COUNT (TapeNum) FROM S-CHeckedOut Temporal generalization of the above query: time-varying count of tape
25、s checked out If now is replaced with a fixed time value, this can be done in SQL in 6 steps and 35 lines! Specifying a key constraint: ALTER TABLE S-CheckedOut ADD PRIMARY KEY (TapeNum) TapeNum is also a key for V-CheckedOut at each point in time It takes 12 line and a complex SQL statement to expr
26、ess this constraint,Querying ,Hence, some 40 temporal query languages have been proposed (most with their own data model), e.g., TSQL2 Simple queries should remain simple: VALIDTIMESELECT COUNT (TapeNum) FROM V-CheckedOut CONSTRAINT temporalkey VALIDTIME UNIQUE TapeNum Early languages based on: rela
27、tional algebra Later: calculus-based, Datalog-based and OO Recent: extensions to SQL,Querying ,Many modeling issues impact the language design, e.g., time stamping tuples or attributes Language design must consider: time-varying nature of data, predicated on temporal values, temporal constructs, sup
28、porting states and/or events, supporting multiple calendars, modification of temporal relations, cursors, views, integrity constraints, handling now, aggregates, schema versioning, periodic data,Querying ,Desired properties of temporal query languages: Temporal upward compatibility: conventional que
29、ries and modifications of temporal relations should act on the current state Pervasive support for sequence queries: that request the history of something, e.g., temporal aggregation above Support for point-based and interval-based view of data Adequate expressive power Ability to be efficiently imp
30、lemented,DBMS Design,Database schemas capturing time-referenced data are complex Two traditional contexts of database design: Data model of DBMS at 3 levels: view, logical, physical (e.g., relational model for the first two) A high-level conceptual design model: ER model Then, mappings bring a conce
31、ptual design into a schema that conforms to the specific implementation data model (e.g., ER to relational mapping) Here: we consider temporal database “logical” and “conceptual” design,Logical Design,Need for guidelines such as formalization guidelines, but conventional normalization concepts are n
32、ot applicable to temporal relational data models A range of temporal normalization concepts have been proposed: temporal dependencies, keys and normal forms Conventional dependencies do not apply: TapeNum does not determine cID, (go through 3 examples) But it should: at any point in time, a tape can
33、 only be checked out by a single customer TapeNum temporally determines cID, but the reverse does not hold,Logical Design ,A temporal relation satisfies a temporal dependency if all its snapshots satisfy the corresponding conventional dependency How to determine snapshots? Timeslice operators: Tempo
34、ral predicate as argument: e.g., contain A time point as parameter: e.g., (tt, vt) Returns snapshot of the relation corresponding to the specified time point, omitting the timestamp attribute Problem: an atemporal approach! which applies to each snapshot of a temporal relation in isolation and hence
35、 fails to account for “temporal” aspects of data,Logical Design ,Consider dependencies and associated normal forms that hold between time points Build in the notion of time granularity into the normalization concepts Not only consider snapshots computed at non-decomposable time points, but also at c
36、oarser granularities: Video rental examples: day as finest granularity, weeks and months may also be considered,Logical Design ,Introducing new concepts that capture the temporal aspects of data and may form the basis for new database design guidelines Most prominent candidate: time patterns Video r
37、ental example: since the set of tapes checked out by a customer changes more frequently that the customers address, they should be stored in separate relations Another candidate: lifespan Attributes with different lifespan (to avoid null values) or with different precision (hour vs. day) should be s
38、tored separately,Conceptual Design,ER diagrams become obscure and cluttered when an attempt is made to capture temporal aspects (see example) CheckedOut relationship should become ternary by introducing an artificial entity set to capture time of rental However, still issues remain: varying rental p
39、rice over time, transaction time inclusion, Some industrial solution: ignore temporal aspects in the ER diagram and supplement it with textual phrases, e.g., “full temporal support” no automatic mapping from ER to model Dozens of temporally enhanced ER models proposed,Conceptual Design ,Give all exi
40、sting ER constructs temporal semantics, similar to “applies to all snapshots” for normalization Does not result in any new syntactical constructs Rules out databases with non-temporal parts: while the syntax of legacy diagrams remain valid their semantics have changed! Devise new notational shorthan
41、d for frequent temporal aspects in ER diagram (e.g., time varying attributes) Both non-temporal and mixed databases can be modeled More difficult to understand,Conceptual Design ,All existing models assume mapping to relational model None tries to map to one of the several time-extended relational m
42、odels Also mapping to emerging models (e.g., SQL3/ORDBMS) are missing.,DBMS Implementation,Integrated approach: internal modules of a DBMS are modified or extended to support time-varying data Efficiency Layered approach: a software layer interposed between the user applications and DBMS that conver
43、ts temporal query language statements to conventional statements Realistic for short and medium term Popular approach: integrated, utilizing timestamping tuples with time intervals,Query Processing,Temporal queries are large and complex Also, the predicates might be temporal, e.g., overlap among two
44、 time intervals Unlike equality predicate in conventional joins, temporal joins require multiple inequality predicates to be examined: two intervals I and j overlap iff st(i) = end(j) and st(j) = end(i) Coalescing of data should be implemented efficiently: interactions among coalescing, duplicate re
45、moval and ordering,Query Processing ,Opportunities for temporal query optimization: Time advances continuously, hence for transaction time, time value used most recently in updates is the largest value used so far natural sorting and clustering: if current and logically deleted tuples are stored sep
46、arately, then Current clustered on st(tt) Deleted clustered on end(tt) Integrity constraint st(j)end(j) Intervals associated with a key value are contiguous in time (end of one interval is the beginning of the other),Implementation of Algebraic Operators,Efficient implementation of temporal selectio
47、n, joins, aggregates, and duplicate elimination temporal index structures Variety of binary temporal joins have been proposed: time-join, time-equijoin, as extensions of nested loop or merge join that exploits orders or local workspace as well as partitioning based joins Also, incremental techniques
48、 for implementing operators on relations capturing transaction time have been discussed Caching the results of previous computations to be reused later (easy to do since the records of updates, I.e., changes to previously cached results, are already contained in a temporal DBMS),Imp. Of Algebraic Op
49、s,Efficient implementation of time-varying aggregates Efficient implementation of coalescing: Sorting the argument relation on the explicit attribute values as well as the valid time Perform the merging in the subsequent scan,Indexing Structures,Similar to spatial index structures can be based on traditional indexes such as B+-tree or multidimensional ones such as R-tree Index structures usually used for selection operators Active research investigation: use index structures for temporal joins, coalescing and aggregates,
