Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Distributed databases


Published on

Published in: Technology
  • Be the first to comment

Distributed databases

  1. 1. Distributed Databases by Chien-Pin Hsu CS157B Section 1 Nov 11, 2004 Dr. Sin-Min Lee
  2. 2. Distributed Database System A distributed database system consists of loosely coupled sites that share no physical component Appears to user as a single system Database systems that run on each site are independent of each other Processing maybe done at a site other than the initiator of request
  3. 3. Homogenous Distributed Database Systems  All sites have identical software  They are aware of each other and agree to cooperate in processing user requests  It appears to user as a single system
  4. 4. An Homogenous Distributed Database Systems example  A distributed system connects three databases: hq, mfg, and sales  An application can simultaneously access or modify the data in several databases in a single distributed environment.
  5. 5. What can we do? A single query from a Manufacturing client on local database mfg can retrieve joined data from the products table on the local database and the dept table on the remote hq database.  For a client application, the location and platform of the databases are transparent.
  6. 6. Makes life easier!! For example, if you are connected to database mfg but want to access data on database hq, creating a synonym on mfg for the remote dept table enables you to issue this query: SELECT * FROM dept In this way, a distributed system gives the appearance of native data access.  Users on mfg do not have to know that the data they access resides on remote databases.
  7. 7. Heterogeneous Distributed Database System  In a heterogeneous distributed database system, at least one of the databases uses different schemas and software.  A database system having different schema may cause a major problem for query processing.  A database system having different software may cause a major problem for transaction processing.
  8. 8. Distributed Data Storage  Replication – System maintains multiple copies of data, stored in different sites, for faster retrieval and fault tolerance.  Fragmentation – Relation is partitioned into several fragments stored in distinct sites Replication and fragmentation can be combined • Relation is partitioned into several fragments: system maintains several identical replicas of each such fragment.
  9. 9. Advantages of Replication Availability: failure of site containing relation r does not result in unavailability of r is replicas exist. Parallelism: queries on r may be processed by several nodes in parallel. Reduced data transfer: relation r is available locally at each site containing a replica of r.
  10. 10. Disadvantages of Replication Increased cost of updates: each replica of relation r must be updated. Increased complexity of concurrency control: concurrent updates to distinct replicas may lead to inconsistent data unless special concurrency control mechanisms are implemented. • One solution: choose one copy as primary copy and apply concurrency control operations on primary copy.
  11. 11. Fragmentation  Data can be distributed by storing individual tables at different sites  Data can also be distributed by decomposing a table and storing portions at different sites – called Fragmentation  Fragmentation can be horizontal or vertical
  12. 12. Why use Fragmentation?  Usage - in general applications use views so it’s appropriate to work with subsets  Efficiency - data stored close to where it is most frequently used  Parallelism - a transaction can divided into several sub-queries to increase degree of concurrency  Security - data more secure - only stored where it is needed Disadvantages: Performance - may be slower Integrity - more difficult
  13. 13. Horizontal Fragmentation Each fragment, Ti , of table T contains a subset of the rows Each tuple of T is assigned to one or more fragments. Horizontal fragmentation is lossless
  14. 14. Horizontal Fragmentation Example  A bank account schema has a relation Account-schema = (branch-name, account-number, balance).  It fragments the relation by location and stores each fragment locally: rows with branch-name = `Hillside` are stored in the Hillside in a fragment
  15. 15. Vertical Fragmentation  Each fragment, Ti, of T contains a subset of the columns, each column is in at least one fragment, and each fragment includes the key: Ti = Πattr_listi (T) T = T1 T2 ….. Tn  All schemas must contain a common candidate key (or superkey) to ensure lossless join property.  A special attribute, the tuple-id attribute may be added to each schema to serve as a candidate key.
  16. 16. Vertical Fragmentation Example A employee-info schema has a relation employee-info schema = (designation, name, Employee-id, salary). It fragments the relation to put information in two tables for security concern.
  17. 17. Commit Protocols  Commit protocols are used to ensure atomicity across sites  Atomicity states that database modifications must follow an “all or nothing” rule.  a transaction which executes at multiple sites must either be committed at all the sites, or aborted at all the sites.
  18. 18. The Two-Phase Commit (2 PC) Protocol What is this?  Two-phase commit is a transaction protocol designed for the complications that arise with distributed resource managers.  Two-phase commit technology is used for hotel and airline reservations, stock market transactions, banking applications, and credit card systems.  With a two-phase commit protocol, the distributed transaction manager employs a coordinator to manage the individual resource managers. The commit process proceeds as follows:
  19. 19. Phase1: Obtaining a Decision Step 1  Coordinator asks all participants to prepare to commit transaction Ti.  Ci adds the records <prepare T> to the log and forces log to stable storage (a log is a file which maintains a record of all changes to the database)  sends prepare T messages to all sites where T executed
  20. 20. Phase1: Making a Decision Step 2  Upon receiving message, transaction manager at site determines if it can commit the transaction  if not: add a record <no T> to the log and send abort T message to Ci  if the transaction can be committed, then: 1). add the record <ready T> to the log 2). force all records for T to stable storage 3). send ready T message to Ci
  21. 21. Phase 2: Recording the Decision  Step 1  T can be committed of Ci received a ready T message from all the participating sites: otherwise T must be aborted.  Step 2  Coordinator adds a decision record, <commit T> or <abort T>, to the log and forces record onto stable storage. Once the record is in stable storage, it cannot be revoked (even if failures occur)  Step 3  Coordinator sends a message to each participant informing it of the decision (commit or abort)  Step 4  Participants take appropriate action locally.
  22. 22. Two-Phase Commit Diagram
  23. 23. Costs and Limitations There have been two performance issues with two phase commit: – If one database server is unavailable, none of the servers gets the updates. – This is correctable through network tuning and correctly building the data distribution through database optimization techniques.