Different levels of Consistency guarantees.pptx

consistency guarantees

1. PNUTS ( Platform for Nimble Universal Table Storage )

2. PNUTS Massively parallel and geographically distributed database system for yahoo’s web applications Hosted Centrally managed Geographically distributed service Automated load balancing and failover

3. Architecture

4. Architecture (contd..) Tablet – group of records Tablet controllers – Mapping of tablets to storage units Routers – directs read/write request to appropriate tablet Storage units – servers Message broker – replication( async ) and consistency

5. A record in PNUTS

6. Record level mastering Based on the write locality characteristic of web workloads One copy of record is designated as master All updates are done to master copy Records are send to non-master using message broker (reliable delivery) Updates on non-master must be forwarded to master before commit

7. Data and query model Data is stored as records in table Flexible schema Attributes can be added on the fly Per record time-line consistency without locks Supports single table query only primary key is used for update and delete Provide flexible access Point access vs range access What is not supported Referential integrity Complex queries like join, group etc

8. A typical web application Consider a social media application where a users can post photos and control access. Consider a scenario - user want to post some party photos but don’t want his parent to see them User action (for n users) U1 : Remove parents from access list U2 : post party photos Consistency model of the underlying system Serializable or eventual?

9. A typical web application Web applications Scalability Response time and geographic scope High availability and fault tolerance Relaxed consistency guarantees Serializable transactions – impractical Eventual consistency – not enough