BASE事务模型高级数据库及大规模存储技术.pptx

高级数据库及大规模存储技术AgendaCAP原理BASE模型一致性问题1.1 ACID语义Atomicity: All or nothing. Consistency: Consistent state of data and transactions. Isolation: Transactions are isolated from each other. Durability: When the transaction is committed, state will be durable.任何存储都可以获得原子性、格里性和持久性，但是我们会一直需要一致性？No！可以通过放弃ACID特性，以达到更高性能和可伸缩性1.2 CAP定理Also known as Brewer’s Theorem by Prof. Eric Brewer, published in 2000 at University of Berkeley.“Of three properties of a shared data system: data consistency, system availability and tolerance to network partitions, only two can be achieved at any given moment.”Proven by Nancy Lynch et al. MIT labs./~brewer/cs262b-2004/PODC-keynote.pdf/~/~brewer/cs262b-2004/PODC-keynote.pdfbrewer/cs262b-2004/PODC-keynote.pdf1.3 CAP语义1.3 CAP 语义Consistency任何节点在同一时刻看到的数据是一样的Availability每个请求都会得到一个是否成功地响应Partial Tolerance由于网络失效，数据划分成跨越网络的片段尽管消息丢失或者系统部分失效，系统依然能够运转 1.3.1 Consistencyread account Xbalance - 100set account X balance = 0set account X balance = 0set account X balance = 0set account X balance = 0DB node 1DB node 2Client 2Client 1 1.3.2 Availabilityread account Xbalance - 100read account Xbalance - 100set account X balance = 0set account X balance = 0set account X balance = 0set account X balance = 0Servers eventually agreebalance is 0.Happy customer walksaway, richer by 200.DB node 1DB node 2Client 2Client 11.3.3 网络失效问题IP网络会面临所有情况TCP网络duplicate、reorderDrop是一种特殊的delay完全的失效检测是不可能的1 /homes/arvind/cs425/doc/fischer.pdf“Impossibility of Distributed Consensus with One Faulty Process”, Fischer, Lynch and Paterson1.4 证明（1）Let’s consider a simple system:Service A writes valuesService B reads valuesValues are replicated between nodesThese are “ideal” systemsBug-free, predictableNode 1V0ANode 2V0B1.4 证明（2）“Sunny day scenario”:A writes a new value V1The value is replicated to node 2B reads the new valueNode 1V1V1V0ANode 2V1V0V1B1.4 证明（3）What happens if the network drops?A writes a new value V1Replication failsB still sees the old valueThe system is inconsistentNode 1V1V1