7. Distributed Database Systems 分布式数据库系统

7.1 Introduction

What is DDB?
A DDB is a collection of correlated data which are spread across a network and managed by a software called DDBMS.
相关数据的集合，相关数据分布在网络的不同节点上，并且由分布式数据库管理系统
Two kinds:

1. Distributed physically, centralized logically (general DDB)
2. Distributed physically, distributed logically too (FDBMS)
   We take the first as main topic in the course.

Features of DDBS

• Distribution
• Correlation
• DDBMS

The advantages of DDBS

• Local autonomy 局部自治性
• Good availiability (because support multi copies) 可用性好
• Good flexibility 灵活
• Low system cost 成本低
• parallel process

The disadvantages of DDBS

• hard to intergrate existing databases
• Too complex

The main problems in DDBS

• Query Optimization (different optimization goal)
• Concurrency control (should consider whole network)
• Recovery mechanism (all sub-transactions must commit or abort simultaneously)

Another problem specially for DDBS

• Data distribution

Data Distribution

• Centralized 集中
• Partitioned 划分
• Replicated 全复制
• Hybird (mix of the above)

Comparison of four strategies

Unit of Data Distribution

1. According to relation (or file), that means non partition 以关系为单位
2. According to Fragments 以裂片为单位

• Horizontal fragmentation: tuple partition 水平分割，元组划分，例如院系划分
• Veritical fragmentation: attribute partition 垂直分割，属性分割
• Mixed fragmentation: both

The criteria of fragmentation 原则

1. Completeness: every tuple or attribute must has its reflection in some fragments 完整性
2. Reconstruction: should be able to reconstruct the original global relation 可重构性
3. Disjointness: for horizontal fragmentation 不相交性

Problem cause by Data Distribution

1. Multi copies’ consistency 多副本的一致性问题
2. Distribution consistency 分布一致性

• Redistribution 重新分布
• Piggybacking 背回

3. Translation of Global Queries to Fragment Queries and Selection of Physical Copies 裂片查询和物理副本选择
4. Design of Database Fragments and Allocation of Fragments 裂片设计

Federated Database

• In practical applications, there are strong requirements for solving the integration of multi existing, distributed and hererogeneous databases.
• The database system in which every member is autonomic and collaborate each other based on negotiation — federated database system
  每个联邦成员都是自治的并且通过协商后互相合作 ，这样的数据库系统称为联邦式数据库
• No global schema in federated database system, every federated member keeps its own data schema
  在联邦式数据库中没有全局模式来让每个成员保持自身的数据模式
• The member negotiate each other to decide respective input/output schema, then, the data sharing relations between each other are established.
• 成员通过协商决定各自的输入和输出模式

The schema structure in federated database System

7.4 Query Optimization in DDBMS

• Optimization goal: minimize the transmission cost on network
  优化目标：减少网上传送量
• Algebra optimization 代数优化
  • Translation of global queries to fragment queries and selection of physical copies 全局关系的查询转化为物理副本的选择
• Query Decomposition
  • Global query plan

An example of global query optimization

Global query optimization may get an execution plan baseed on cost estimation, such as:

1. send R2 to site 1, R’
2. execute on site1:
   Select *
   From R1, R’
   WHERE R1.a = R’.b;

7.5 Recovery Mechanisim in DDBMS

• The basic principle is the same as that in centralized DBMS
• Distributed transactions: the key of distributed transaction management is how to assure all sub-transactions either commit together or abort together.
  分布式事务：
• Realize the sub-transactions’ harmony with each other relies on communication, while the communication is not reliable.
• Two phase commitment protocol 两段提交协议
• Combination of failures 故障的组合问题

7.6 Concurrency Control in DDBMS

• The basic principle is the same as that in centralized DBMS， demand concurrent transactions to scheduled serializably
• Because of multi copies, need locking globally
• Communication overhead
• Global deadlock