1. 1 Lecture 8 Distributed Data Bases: Replication and Fragmentation

2. 2 Overview Last lecture: Saw difficulty in handling logical relationships between distributed information. Potential solutions such as federated DDBMS. This week: Look at an area where distributed data bases are extensively used: Replication. For back-up –for improving reliability of service –e.g. mirror site.

3. 3 Strategies for Data Allocation (1) Centralised Single data base, users distributed across network. High communication costs: All data access by users over network; No local references. Storage costs: No duplication, so minimal. Low reliability and low availability: Failure of central site prevents access to entire data base system. Performance: Likely to be unsatisfactory.

4. 4 Strategies for Data Allocation (2) Fragmented Data base distributed by fragments (disjoint views). Low communication costs: Fragments located near their main users (if well designed). Storage costs: No duplication, so minimal. Reliability and availability vary depending on failed site: Failure of one part loses fragments situated there; Other fragments continue to be available. Performance: Likely to be satisfactory – better than centralised, since less network traffic.

5. 5 Strategies for Data Allocation (3) Complete Replication Data base completely copied to each site. Communication costs: Low for reads, high for updates –need to propagate updates through system. High Storage costs: Complete duplication. High reliability and high availability: Can switch from failed site to another. Performance: Good for reads; Potentially poor for updates, because of propagation.

6. 6 Strategies for Data Allocation (4) Selective Replication Fragments are selectively replicated. Communication costs: Low (if well designed). Storage costs Duplication of some fragments means that it is not minimal, but less than with complete replication. Reliability and availability vary depending on failed site: Failure of one part loses fragments situated (only) there; Other fragments continue to be available. Performance: Likely to be satisfactory – better than centralised, since less network traffic.

7. 7 Fragmentation – Further Details A fragment is a view on a table. Two main types: Horizontal (classification by value) Sub-set of tuples obtained by RESRICT operation (algebra) or WHERE clause (SQL). Vertical (classification by property) Sub-set of columns obtained by PROJECT operation (algebra) or SELECT clause (SQL).

8. 8 Other Forms of Fragmentation Mixed (classification by both value and property) Both horizontal and vertical fragmentation are used to obtain a single fragment. Derived (association) An expression such as a join connects the fragments. None The whole of a table appears, unchanged, in a view.

9. 9 Why Fragment? Most applications use only part of the data in a table. To minimise network traffic, do not send more data to any site than is strictly necessary. Data not required by an application is not visible to it, enhancing security.

10. 10 Factors against Fragmentation Performance May be affected adversely by the need for some applications to reconstruct fragments into larger units. Integrity More difficult to control, with dependencies possibly scattered across fragments.

11. 11 3 Rules for Fragmentation Rule 1: Completeness If a table T is decomposed into fragments: Every value found in T must be contained in at least one of the fragments –so no loss of data in fragmentation –otherwise we lose data.

12. 12 Rule 2: Reconstruction It must be possible to reconstruct T from the fragments using a relational operation (typically a natural join) –so functional dependencies are preserved –otherwise decomposition into fragments is lossy. 3 Rules for Fragmentation

13. 13 Rule 3: Disjointness A data item may not appear in more than one fragment unless it is a component of a Primary Key. This avoids duplication and potential inconsistency –although transactions should avoid the latter. Primary Key duplication allows reconstructions to be made. 3 Rules for Fragmentation

14. 14 Strategy for Designing a Partially Replicated Distributed Data Base – 1 Design a global data base using standard methodology. Examine the regional distribution of the business. What data is needed by each part of the business? Some data is only used locally (not exported, as in a Federated DDBMS). Some data is mostly used locally.

15. 15 Transactions give many clues to ideal placement of fragments a transaction will perform slowly if it requires data from different sites, unless the network connecting them is very fast. a transaction performing much replication of updates will perform slowly if there is frequent contention for resources (locking). frequently used transactions should be optimised; infrequently used ones can be ignored. Strategy for Designing a Partially Replicated Distributed Data Base – 2

16. 16 Decide on which relations are not to be fragmented. They will normally be replicated everywhere: –for ease of updating and to maintain integrity. Fragment remaining relations to suit: locality; transactions. Strategy for Designing a Partially Replicated Distributed Data Base – 3

17. 17 Transparencies in DDBMS Transparency hides low-level details (often details of implementation) from the user. Four main types: Distribution; Transaction; Performance; DBMS.

18. 18 Distribution Transparency The DDB is perceived by the user as a single, logical unit even though the data is: distributed over several sites; fragmented in various ways.

19. 19 Significance of Full Distribution Transparency The highest form of Distribution Transparency is termed Fragmentation Transparency. Users do not need to know anything about the distribution techniques. Users address the global schema in queries. Users will not, however, understand why some queries take longer than others.

20. 20 Reduced Forms of Distribution Transparency Location Transparency Users need to know about fragmentation but not about placements at sites. Users do not need to know what replications exist. Local Mapping Transparency The most limited transparency. Users need to know about fragmentation and sites.

21. 21 Transaction Transparency Ensures that all transactions maintain the DDB’s integrity and consistency. Each transaction is divided into sub- transactions: one sub-transaction for each site; sub-transactions usually executed in parallel –hence gains in efficiency More complicated than in centralised system.

22. 22 Forms of Transaction Transparency Concurrency Transparency All concurrent transactions (centralised and distributed) execute independently. The DDBMS must ensure that: each sub-transaction is executed in the normal spirit of transactions (ACID); the sub-transactions as a whole, forming one transaction, are executed ACID-style; the mixture of sub-transactions and whole transactions is executed ACID-style.

23. 23 Transactions – Problems with Replication Failure Transparency Users are unaware of problems encountered during transaction execution, e.g. If, say, 6 copies of a data item (at 6 sites) need to be updated: Problems if only 5 are currently reachable. Need to delay COMMIT until all sites processed –otherwise inconsistent data –unless we allow a delayed asynchronous update.

24. 24 Performance Transparency Requires the DDBMS to determine the most cost-effective way to handle a request: which fragment to use; (if replicated) which copy of a fragment to use; which site to use; Avoidance of any performance degradation compared with a centralised system.

25. 25 DBMS Transparency Hides knowledge of which DBMS is being used. The most difficult transparency of all –particularly with heterogeneous models. See problems highlighted in Lecture 7: Global Schema Integration; Federated Data Bases; Multi-Data-Base Languages.

26. 26 Replication Servers Copying and maintenance of data on multiple servers. Replication – the process of generating and reproducing multiple copies of data at one or more sites. Servers – provide the file resources – the distributed data base.

27. 27 Benefits of Replication Increased reliability. Better data availability. Potential for better performance (with good design). “Warm stand-by” – as in mirror site, shadowing actions of main site and taking over if main site crashes.

28. 28 Timing of Replication Synchronous Immediate, according to some common signal such as time. Ideal as it ensures immediate consistency. Assumes availability of all sites. Asynchronous Independently with delays ranging from a few seconds to several days. Immediate consistency is not achieved. More flexible, since at any one time not all sites need be available.

29. 29 Types of Data Replicated Across heterogeneous data models: Mapping required (hard). Object replication: More varied than just base tables. Also auxiliary structures such as indices. Stored procedures and functions. Scalability: No volume restrictions.

30. 30 Replication Administration Subscription Mechanism Allows a permitted user to subscribe to replicated data / objects. Initialisation Mechanism Allows for the initialisation of a target replication.

31. 31 Ownership of Replicated Data (1) Master / Slave Master Site Primary owner of replicated data. Has sole right to change data. Publish and subscribe procedure. Asynchronous replication as slave sites receive copies of the data. Slave site Receive read-only data from master site. Slaves can be used as mobile clients.

32. 32 Work-Flow Ownership Flexible master designation. Dynamic ownership model. Right to update data moves along the chain of command (replicating sites). E.g. as an order is processed the master right moves to each department in turn. Ownership of Replicated Data (2)

33. 33 Update Anywhere Peer-to-peer model. Multiple sites can update data. Conflict resolution required. More complex implementation. Ownership of Replicated Data (3)

34. 34 Distribution and Replication in Oracle 9i Materialised Views –formerly known as Snapshots. Views are updated by Refresh mechanism; Variable frequency to suit application: Fast – based on identified changes; Complete – replaces all existing data; Force – tries Fast – if not possible, does Complete.

35. 35 Oracle 9i Transparency Supports Site (Location) Transparency. Does not support Fragmentation Transparency.

36. 36 Summary of Distributed DBMS An area under development to improve: Availability of data; Overall reliability of system; Performance (through good design). However, disadvantages remain: Implementation can be complex (expensive). Heterogeneity in models is poorly handled. Main use today is for replicating data.