ClusteringUseCases

Overview

• Nothing is really 24/7 production ready without advanced (4-5,6 9s availability)
• Without substantial expert DBA staff
• Clustering != replication

3 major use cases for clustering

1. Transactional (least change in the next 3 years)

• 99.9999 uptime :)
• 60k Write transactions per second
• -> needs to scale to 200,000 writes per second
• 5,10,20 servers
• 0% data loss is acceptable (committed transactions)
  • will sacrifice scalability, or flexibility so that we do not lose any data

2. Analytic use case

• Looking Parallelizing queries --> execution time
• Maximize amount of CPU + I/O
• (Greenplum, Aster)
• 100s of nodes
• No transactions
• in last 3 years, 11 data warehousing startups (7 using postgres)

3. Online

• No products exist to address this
• Web -> mobile services (android, iphone, symbian/Qt)
• Scale from 100k -> 1 million users -> 1 billion users
• Looking for Scalability, read and write
  • Read scalability through caching (?) .. but not everything can be put here
• Rapid Deployment
• Loose xtns -> needs them for SOME things, but MOST of online services, txn is not important
  • some amount of data loss is acceptable :)
• user-centered providing of data
  • eventually consistent

JAN:

• Geographic distribution
• may be online use case

See The Three Database Clustering Users for a longer version of the above