Chapter 25 Distributed DBMSs - Advanced Concepts Pearson Education © 2009
Chapter 25 - Objectives u Distributed transaction management. u Distributed concurrency control. u Distributed deadlock detection. u Distributed recovery control. u Distributed integrity control. u X/OPEN DTP standard. u Distributed query optimization. u Oracle’s DDBMS functionality. Pearson Education © 2009 2
Distributed Transaction Management u Distributed transaction accesses data stored at more than one location. u Divided into a number of sub-transactions, one for each site that has to be accessed, represented by an agent. u Indivisibility of distributed transaction is still fundamental to transaction concept. u DDBMS must also ensure indivisibility of each sub-transaction. Pearson Education © 2009 3
Distributed Transaction Management u Thus, DDBMS must ensure: – synchronization of subtransactions with other local transactions executing concurrently at a site; – synchronization of subtransactions with global transactions running simultaneously at same or different sites. u Global transaction manager (transaction coordinator) at each site, to coordinate global and local transactions initiated at that site. Pearson Education © 2009 4
Coordination of Distributed Transaction Pearson Education © 2009 5
Distributed Locking u Look at four schemes: – Centralized Locking. – Primary Copy 2 PL. – Distributed 2 PL. – Majority Locking. Pearson Education © 2009 6
Centralized Locking Single site that maintains all locking information. u One lock manager for whole of DDBMS. u Local transaction managers involved in global transaction request and release locks from lock manager. u Or transaction coordinator can make all locking requests on behalf of local transaction managers. u Advantage - easy to implement. u Disadvantages - bottlenecks and lower reliability. u Pearson Education © 2009 7
Primary Copy 2 PL u Lock managers distributed to a number of sites. u Each lock manager responsible for managing locks for set of data items. u For replicated data item, one copy is chosen as primary copy, others are slave copies u Only need to write-lock primary copy of data item that is to be updated. u Once primary copy has been updated, change can be propagated to slaves. Pearson Education © 2009 8
Primary Copy 2 PL u Disadvantages - deadlock handling is more complex; still a degree of centralization in system. u Advantages - lower communication costs and better performance than centralized 2 PL. Pearson Education © 2009 9
Distributed 2 PL u Lock managers distributed to every site. u Each lock manager responsible for locks for data at that site. u If data not replicated, equivalent to primary copy 2 PL. u Otherwise, implements a Read-One-Write-All (ROWA) replica control protocol. Pearson Education © 2009 10
Distributed 2 PL u Using ROWA protocol: – Any copy of replicated item can be used for read. – All copies must be write-locked before item can be updated. u Disadvantages - deadlock handling more complex; communication costs higher than primary copy 2 PL. Pearson Education © 2009 11
Majority Locking u Extension of distributed 2 PL. u To read or write data item replicated at n sites, sends a lock request to more than half the n sites where item is stored. u Transaction cannot proceed until majority of locks obtained. u Overly strong in case of read locks. Pearson Education © 2009 12
Distributed Timestamping u Objective is to order transactions globally so older transactions (smaller timestamps) get priority in event of conflict. u In distributed environment, need to generate unique timestamps both locally and globally. u System clock or incremental event counter at each site is unsuitable. u Concatenate local timestamp with a unique site identifier: <local timestamp, site identifier>. Pearson Education © 2009 13
Distributed Timestamping u Site identifier placed in least significant position to ensure events ordered according to their occurrence as opposed to their location. u To prevent a busy site generating larger timestamps than slower sites: – Each site includes their timestamps in messages. – Site compares its timestamp with timestamp in message and, if its timestamp is smaller, sets it to some value greater than message timestamp. Pearson Education © 2009 14
Distributed Deadlock u More complicated if lock management is not centralized. u Local Wait-for-Graph (LWFG) may not show existence of deadlock. u May need to create GWFG, union of all LWFGs. u Look at three schemes: – Centralized Deadlock Detection. – Hierarchical Deadlock Detection. – Distributed Deadlock Detection. Pearson Education © 2009 15
Example - Distributed Deadlock · · · T 1 initiated at site S 1 and creating agent at S 2, T 2 initiated at site S 2 and creating agent at S 3, T 3 initiated at site S 3 and creating agent at S 1. Time S 1 S 2 t 1 read_lock(T 1, x 1) write_lock(T 2, y 2) S 3 read_lock(T 3, z 3) t 2 write_lock(T 1, y 1) write_lock(T 2, z 2) t 3 write_lock(T 3, x 1) write_lock(T 1, y 2) Pearson Education © 2009 write_lock(T 2, z 3) 16
Example - Distributed Deadlock Pearson Education © 2009 17
Centralized Deadlock Detection u Single site appointed deadlock detection coordinator (DDC). u DDC has responsibility for constructing and maintaining GWFG. u If one or more cycles exist, DDC must break each cycle by selecting transactions to be rolled back and restarted. Pearson Education © 2009 18
Hierarchical Deadlock Detection u Sites are organized into a hierarchy. u Each site sends its LWFG to detection site above it in hierarchy. u Reduces dependence on centralized detection site. Pearson Education © 2009 19
Hierarchical Deadlock Detection Pearson Education © 2009 20
Distributed Deadlock Detection u Most well-known method developed by Obermarck (1982). u An external node, Text, is added to LWFG to indicate remote agent. u If a LWFG contains a cycle that does not involve Text, then site and DDBMS are in deadlock. Pearson Education © 2009 21
Distributed Deadlock Detection u Global deadlock may exist if LWFG contains a cycle involving Text. u To determine if there is deadlock, the graphs have to be merged. u Potentially more robust than other methods. Pearson Education © 2009 22
Distributed Deadlock Detection Pearson Education © 2009 23
Distributed Deadlock Detection S 1: S 2: S 3: Text T 3 T 1 Text T 1 T 2 Text T 2 T 3 Text u Transmit LWFG for S 1 to the site for which transaction T 1 is waiting, site S 2. u LWFG at S 2 is extended and becomes: S 2: Text T 3 T 1 T 2 Text Pearson Education © 2009 24
Distributed Deadlock Detection u Still contains potential deadlock, so transmit this WFG to S 3: Text T 3 T 1 T 2 T 3 Text u GWFG contains cycle not involving Text, so deadlock exists. Pearson Education © 2009 25
Distributed Deadlock Detection u Four types of failure particular to distributed systems: – Loss of a message. – Failure of a communication link. – Failure of a site. – Network partitioning. u Assume first are handled transparently by DC component. Pearson Education © 2009 26
Distributed Recovery Control u DDBMS is highly dependent on ability of all sites to be able to communicate reliably with one another. u Communication failures can result in network becoming split into two or more partitions. u May be difficult to distinguish whether communication link or site has failed. Pearson Education © 2009 27
Partitioning of a network Pearson Education © 2009 28
Two-Phase Commit (2 PC) u Two phases: a voting phase and a decision phase. u Coordinator asks all participants whether they are prepared to commit transaction. – If one participant votes abort, or fails to respond within a timeout period, coordinator instructs all participants to abort transaction. – If all vote commit, coordinator instructs all participants to commit. u All participants must adopt global decision. Pearson Education © 2009 29
Two-Phase Commit (2 PC) u If participant votes abort, free to abort transaction immediately u If participant votes commit, must wait for coordinator to broadcast global-commit or global -abort message. u Protocol assumes each site has its own local log and can rollback or commit transaction reliably. u If participant fails to vote, abort is assumed. u If participant gets no vote instruction from coordinator, can abort. Pearson Education © 2009 30
2 PC Protocol for Participant Voting Commit Pearson Education © 2009 31
2 PC Protocol for Participant Voting Abort Pearson Education © 2009 32
2 PC Termination Protocols u Invoked whenever a coordinator or participant fails to receive an expected message and times out. Coordinator u Timeout in WAITING state – Globally abort transaction. u Timeout in DECIDED state – Send global decision again to sites that have not acknowledged. Pearson Education © 2009 33
2 PC - Termination Protocols (Participant) u Simplest termination protocol is to leave participant blocked until communication with the coordinator is re-established. Alternatively: u Timeout in INITIAL state – Unilaterally abort transaction. u Timeout in the PREPARED state – Without more information, participant blocked. – Could get decision from another participant. Pearson Education © 2009 34
State Transition Diagram for 2 PC (a) coordinator; (b) participant Pearson Education © 2009 35
2 PC Recovery Protocols u Action to be taken by operational site in event of failure. Depends on what stage coordinator or participant had reached. Coordinator Failure u Failure in INITIAL state – Recovery starts commit procedure. u Failure in WAITING state – Recovery restarts commit procedure. Pearson Education © 2009 36
2 PC Recovery Protocols (Coordinator Failure) u Failure in DECIDED state – On restart, if coordinator has received all acknowledgements, it can complete successfully. Otherwise, has to initiate termination protocol discussed above. Pearson Education © 2009 37
2 PC Recovery Protocols (Participant Failure) u Objective to ensure that participant on restart performs same action as all other participants and that this restart can be performed independently. u Failure in INITIAL state – Unilaterally abort transaction. u Failure in PREPARED state – Recovery via termination protocol above. u Failure in ABORTED/COMMITTED states – On restart, no further action is necessary. Pearson Education © 2009 38
2 PC Topologies Pearson Education © 2009 39
Three-Phase Commit (3 PC) u 2 PC is not a non-blocking protocol. u For example, a process that times out after voting commit, but before receiving global instruction, is blocked if it can communicate only with sites that do not know global decision. u Probability of blocking occurring in practice is sufficiently rare that most existing systems use 2 PC. Pearson Education © 2009 40
Three-Phase Commit (3 PC) u Alternative non-blocking protocol, called threephase commit (3 PC) protocol. u Non-blocking for site failures, except in event of failure of all sites. u Communication failures can result in different sites reaching different decisions, thereby violating atomicity of global transactions. u 3 PC removes uncertainty period for participants who have voted commit and await global decision. Pearson Education © 2009 41
Three-Phase Commit (3 PC) u Introduces third phase, called pre-commit, between voting and global decision. u On receiving all votes from participants, coordinator sends global pre-commit message. u Participant who receives global pre-commit, knows all other participants have voted commit and that, in time, participant itself will definitely commit. Pearson Education © 2009 42
State Transition Diagram for 3 PC (a) coordinator; (b) participant Pearson Education © 2009 43
3 PC Protocol for Participant Voting Commit 44 Pearson Education © 2009
3 PC Termination Protocols (Coordinator) u Timeout in WAITING state – Same as 2 PC. Globally abort transaction. u Timeout in PRE-COMMITTED state – Write commit record to log and send GLOBAL -COMMIT message. u Timeout in DECIDED state – Same as 2 PC. Send global decision again to sites that have not acknowledged. Pearson Education © 2009 45
3 PC Termination Protocols (Participant) u Timeout in INITIAL state – Same as 2 PC. Unilaterally abort transaction. u Timeout in the PREPARED state – Follow election protocol to elect new coordinator. u Timeout in the PRE-COMMITTED state – Follow election protocol to elect new coordinator. Pearson Education © 2009 46
3 PC Recovery Protocols (Coordinator Failure) u Failure in INITIAL state – Recovery starts commit procedure. u Failure in WAITING state – Contact other sites to determine fate of transaction. u Failure in PRE-COMMITTED state – Contact other sites to determine fate of transaction. u Failure in DECIDED state – If all acknowledgements in, complete transaction; otherwise initiate termination protocol above. Pearson Education © 2009 47
3 PC Recovery Protocols (Participant Failure) u Failure in INITIAL state – Unilaterally abort transaction. u Failure in PREPARED state – Contact other sites to determine fate of transaction. u Failure in PRE-COMMITTED state – Contact other sites to determine fate of transaction. u Failure in ABORTED/COMMITTED states – On restart, no further action is necessary. Pearson Education © 2009 48
3 PC Termination Protocol After New Coordinator u 1. 2. 3. 4. Newly elected coordinator will send STATEREQ message to all participants involved in election to determine how best to continue. If some participant has aborted, then abort. If some participant has committed, then commit. If all participants are uncertain, then abort. If some participant is in PRE-COMMIT, then commit. To prevent blocking, send PRECOMMIT and after acknowledgements, send GLOBAL-COMMIT. Pearson Education © 2009 49
Network Partitioning u If data is not replicated, can allow transaction to proceed if it does not require any data from site outside partition in which it is initiated. u Otherwise, transaction must wait until sites it needs access to are available. u If data is replicated, procedure is much more complicated. Pearson Education © 2009 50
Identifying Updates Pearson Education © 2009 51
Identifying Updates u Successfully completed update operations by users in different partitions can be difficult to observe. u In P 1, transaction withdrawn £ 10 from account and in P 2, two transactions have each withdrawn £ 5 from same account. u At start, both partitions have £ 100 in balx, and on completion both have £ 90 in balx. u On recovery, not sufficient to check value in balx and assume consistency if values same. Pearson Education © 2009 52
Maintaining Integrity Pearson Education © 2009 53
Maintaining Integrity Successfully completed update operations by users in different partitions can violate constraints. u Have constraint that account cannot go below £ 0. u In P 1, withdrawn £ 60 from account and in P 2, withdrawn £ 50. u At start, both partitions have £ 100 in balx, then on completion one has £ 40 in balx and other has £ 50. u Importantly, neither has violated constraint. u On recovery, balx is –£ 10, and constraint violated. u Pearson Education © 2009 54
Network Partitioning u Processing in partitioned network involves tradeoff in availability and correctness. u Correctness easiest to provide if no processing of replicated data allowed during partitioning. u Availability maximized if no restrictions placed on processing of replicated data. u In general, not possible to design non-blocking commit protocol for arbitrarily partitioned networks. Pearson Education © 2009 55
X/OPEN DTP Model u Open Group is vendor-neutral consortium whose mission is to cause creation of viable, global information infrastructure. u Formed by merge of X/Open and Open Software Foundation. u X/Open established DTP Working Group with objective of specifying and fostering appropriate APIs for TP. u Group concentrated on elements of TP system that provided the ACID properties. Pearson Education © 2009 56
X/OPEN DTP Model u X/Open DTP standard that emerged specified three interacting components: – an application, – a transaction manager (TM), – a resource manager (RM). Pearson Education © 2009 57
X/OPEN DTP Model Any subsystem that implements transactional data can be a RM, such as DBMS, transactional file system or session manager. u TM responsible for defining scope of transaction, and for assigning unique ID to it. u Application calls TM to start transaction, calls RMs to manipulate data, and calls TM to terminate transaction. u TM communicates with RMs to coordinate transaction, and TMs to coordinate distributed transactions. u Pearson Education © 2009 58
X/OPEN DTP Model - Interfaces u Application may use TX interface to communicate with a TM. u TX provides calls that define transaction scope, and whether to commit/abort transaction. u TM communicates transactional information with RMs through XA interface. u Finally, application can communicate directly with RMs through a native API, such as SQL or ISAM. Pearson Education © 2009 59
X/OPEN DTP Model Interfaces Pearson Education © 2009 60
X/OPEN Interfaces in Distributed Environment Pearson Education © 2009 61
Distributed Query Optimization 62 Pearson Education © 2009
Distributed Query Optimization u Query decomposition: takes query expressed on global relations and performs partial optimization using centralized QO techniques. Output is some form of RAT based on global relations. u Data localization: takes into account how data has been distributed. Replace global relations at leaves of RAT with their reconstruction algorithms. Pearson Education © 2009 63
Distributed Query Optimization u Global optimization: uses statistical information to find a near-optimal execution plan. Output is execution strategy based on fragments with communication primitives added. u Local optimization: Each local DBMS performs its own local optimization using centralized QO techniques. Pearson Education © 2009 64
Data Localization u In QP, represent query as R. A. T. and, using transformation rules, restructure tree into equivalent form that improves processing. u In DQP, need to consider data distribution. u Replace global relations at leaves of tree with their reconstruction algorithms - RA operations that reconstruct global relations from fragments: – For horizontal fragmentation, reconstruction algorithm is Union; – For vertical fragmentation, it is Join. Pearson Education © 2009 65
Data Localization u Then use reduction techniques to generate simpler and optimized query. u Consider reduction techniques for following types of fragmentation: – Primary horizontal fragmentation. – Vertical fragmentation. – Derived fragmentation. Pearson Education © 2009 66
Reduction for Primary Horizontal Fragmentation u If selection predicate contradicts definition of fragment, this produces empty intermediate relation and operations can be eliminated. u For join, commute join with union. u Then examine each individual join to determine whethere any useless joins that can be eliminated from result. u. A useless join exists if fragment predicates do not overlap. Pearson Education © 2009 67
Example 25. 2 Reduction for PHF SELECT * FROM Branch b, Property. For. Rent p WHERE b. branch. No = p. branch. No AND p. type = ‘Flat’; P 1: branch. No=‘B 003’ type=‘House’ (Property. For. Rent) P 2: branch. No=‘B 003’ type=‘Flat’ (Property. For. Rent) P 3: branch. No!=‘B 003’ (Property. For. Rent) B 1: branch. No=‘B 003’ (Branch) B 2: branch. No!=‘B 003’ (Branch) Pearson Education © 2009 68
Example 25. 2 Reduction for PHF Pearson Education © 2009 69
Example 25. 2 Reduction for PHF Pearson Education © 2009 70
Example 25. 2 Reduction for PHF Pearson Education © 2009 71
Reduction for Vertical Fragmentation u Reduction for vertical fragmentation involves removing those vertical fragments that have no attributes in common with projection attributes, except the key of the relation. Pearson Education © 2009 72
Example 25. 3 Reduction for Vertical Fragmentation SELECT f. Name, l. Name FROM Staff; S 1: staff. No, position, sex, DOB, salary(Staff) S 2: staff. No, f. Name, l. Name, branch. No (Staff) Pearson Education © 2009 73
Example 25. 3 Reduction for Vertical Fragmentation Pearson Education © 2009 74
Reduction for Derived Fragmentation u Use transformation rule that allows join and union to be commuted. u Using knowledge that fragmentation for one relation is based on the other and, in commuting, some of the partial joins should be redundant. Pearson Education © 2009 75
Example 25. 4 Reduction for Derived Fragmentation SELECT * FROM Branch b, Client c WHERE b. branch. No = c. branch. No AND b. branch. No = ‘B 003’; B 1 = branch. No=‘B 003’ (Branch) B 2 = branch. No!=‘B 003’ (Branch) Ci = Client branch. No Bi i = 1, 2 Pearson Education © 2009 76
Example 25. 4 Reduction for Derived Fragmentation Pearson Education © 2009 77
Global Optimization u Objective of this layer is to take the reduced query plan for the data localization layer and find a near-optimal execution strategy. u In distributed environment, speed of network has to be considered when comparing strategies. u If know topology is that of WAN, could ignore all costs other than network costs. u LAN typically much faster than WAN, but still slower than disk access. Pearson Education © 2009 78
Global Optimization u Cost model could be based on total cost (time), as in centralized DBMS, or response time. Latter uses parallelism inherent in DDBMS. Pearson Education © 2009 79
Global Optimization – R* uses a cost model based on total cost and static query optimization. u Like centralized System R optimizer, algorithm is based on an exhaustive search of all join orderings, join methods (nested loop or sortmerge join), and various access paths for each relation. u When Join is required involving relations at different sites, R* selects the sites to perform Join and method of transferring data between sites. Pearson Education © 2009 80
Global Optimization – R* u For a Join of R and S with R at site 1 and S at site 2, there are three candidate sites: – site 1, where R is located; – site 2, where S is located; – some other site (e. g. , site of relation T, which is to be joined with join of R and S). Pearson Education © 2009 81
Global Optimization – R* u In R*, there are 2 methods for transferring data: 1. Ship whole relation 2. Fetch tuples as needed. u u First method incurs a larger data transfer but fewer message then second. R* considers only the following methods: 1. Nested loop, ship whole outer relation to site of inner. 2. Sort-merge, ship whole inner relation to site of outer. 3. Nested loop, fetch tuples of inner relation as needed for each tuple of outer relation. 4. Sort-merge, fetch tuples of inner relation as needed for each tuple of outer relation. 5. Ship both relations to third site. Pearson Education © 2009 82
Global Optimization – SDD-1 u Based on an earlier method known as “hill climbing”, a greedy algorithm that starts with an initial feasible solution which is then iteratively improved. u Modified to make use of Semijoin to reduce cardinality of join operands. u Like R*, SDD-1 optimizer minimizes total cost, although unlike R* it ignores local processing costs and concentrates on communication message size. u Like R*, query processing timing used is static. Pearson Education © 2009 83
Global Optimization – SDD-1 u Based on concept of “beneficial Semijoins”. u Communication cost of Semijoin is simply cost of transferring join attribute of first operand to site of second operand. u “Benefit” of Semijoin is taken as cost of transferring irrelevant tuples of first operand, which Semijoin avoids. Pearson Education © 2009 84
Global Optimization – SDD-1 u Phase 1 – Initialization: Perform all local reductions using Selection and Projection. Execute Semijoins within same site to reduce sizes of relations. Generate set of all beneficial Semijoins across sites (Semijoin is beneficial if its cost is less than its benefit). u Phase 2 – Selection of beneficial Semijoins: Iteratively select most beneficial Semijoin from set generated and add it to execution strategy. After each iteration, update database statistics to reflect incorporation of the Semijoin and update the set with new beneficial Semijoins. Pearson Education © 2009 85
Global Optimization – SDD-1 Phase 3 – Assembly site selection: Select, among all sites, site to which transmission of all relations incurs a minimum cost. Choose site containing largest amount of data after reduction phase so that sum of the amount of data transferred from other sites will be minimum. u Phase 4 – Postoptimization: Discard useless Semijoins; e. g. if R resides in assembly site and R is due to be reduced by Semijoin, but is not used to reduce other relations after Semijoin, then since R need not be moved to another site during assembly phase, Semijoin on R is useless and can be discarded. u Pearson Education © 2009 86
Oracle’s DDBMS Functionality u Oracle does not support type of fragmentation discussed previously, although DBA can distribute data to achieve similar effect. u Thus, fragmentation transparency is not supported although location transparency is. u Discuss: – – – connectivity global database names and database links transactions referential integrity heterogeneous distributed databases Distributed QO. Pearson Education © 2009 87
Connectivity – Oracle Net Services u Oracle Net Services supports communication between clients and servers. u Enables both client-server and server-server communication across any network, supporting both distributed processing and distributed DBMS capability. u Also responsible for translating any differences in character sets or data representation that may exist at operating system level. Pearson Education © 2009 88
Global Database Names u Unique name given to each distributed database. u Formed by prefixing the database’s network domain name with the local database name. u Domain name follows standard Internet conventions, with levels separated by dots ordered from leaf to root, left to right. 89 Pearson Education © 2009
Database Links u Used to build distributed databases. u Defines a communication path from one Oracle database to another (possibly non-Oracle) database. u Acts as a type of remote login to remote database. CREATE PUBLIC DATABASE LINK RENTALS. GLASGOW. NORTH. COM; SELECT * FROM Staff@RENTALS. GLASGOW. NORTH. COM; UPDATE Staff@RENTALS. GLASGOW. NORTH. COM SET salary = salary*1. 05; Pearson Education © 2009 90
Types of Transactions u Remote SQL statements: Remote query selects data from one or more remote tables, all of which reside at same remote node. Remote update modifies data in one or more tables, all of which are located at same remote node. u Distributed SQL statements: Distributed query retrieves data from two or more nodes. Distributed update modifies data on two or more nodes. u Remote transactions: Contains one or more remote statements, all of which reference a single remote node. Pearson Education © 2009 91
Types of Transactions u Distributed transactions: Includes one or more statements that, individually or as a group, update data on two or more distinct nodes of a distributed database. Oracle ensures integrity of distributed transactions using 2 PC. Pearson Education © 2009 92
Referential Integrity u Oracle does not permit declarative referential integrity constraints to be defined across databases. u However, parent-child table relationships across databases can be maintained using triggers. Pearson Education © 2009 93
Heterogeneous Distributed Databases u Here one of the local DBMSs is not Oracle. u Oracle Heterogeneous Services and a non-Oracle system-specific agent can hide distribution and heterogeneity. u Can be accessed through: – transparent gateways – generic connectivity. Pearson Education © 2009 94
Transparent Gateways 95 Pearson Education © 2009
Generic Connectivity 96 Pearson Education © 2009
Oracle Distributed Query Optimization u. A distributed query is decomposed by the local Oracle DBMS into a number of remote queries, which are sent to remote DBMS for execution. u Remote DBMSs execute queries and send results back to local node. u Local node then performs any necessary postprocessing and returns results to user. u Only necessary data from remote tables are extracted, thereby reducing amount of data that needs to be transferred. Pearson Education © 2009 97
Скачать презентацию Chapter 25 Distributed DBMSs — Advanced Concepts Pearson
5bbed2aad630cfcf667bb4118d45d3fc.ppt