Presentation: Gary Dusbabek (Rackspace) on Cassandra

A presentation about Cassandra given by Rackspace’ Gary Dusbabek (@gdusbabek):

My notes:

What problems does it solve?

• Reliability at scale
  • No Single point of failure (all nodes are identifical)
• Simple scaling
  • linear
• High write thoughput
• Large data sets

What problems can’t it solve?

• No flexible indices
• No querying on non PK values
• Not good for binary data (>64mb) unless you chunck
• Row contents must fit in available memory

Concepts: CAP

• Cassandra chooses A and P but allows them to be tunable to have more C

Data Model

• Keyspace contains column families
• ColumnFamily:
  • Standard or Super
  • Two levels of indexes (key and column names)

Data Model

• Column and subcolumn sorting
• Specify your own comparator:
  • TimeUUID
  • Lexical UUID
  • UTF8
  • Bytes
  • CreateYourOwn

Inserting: Writes

• Commit log for durability
• Memtable - no disk access (no reads or seeks)
• Sstables are final (become read only)
  • Index
  • Bloom filter
  • Raw data
• Atomic within a ColumnFamily
• Bottom line: FAST!!

Note: make sure to check the slide for a nice visual description of Cassandra write operation. You should check also the Cassandra Write operation performance explained for more details.

Querying: Overview

• But secondary indices are being worked on (see ☞ CASSANDRA-749)

Querying: Reads

• Not as fast as writes
• Read repair when out of sync
• New in 0.6:
  • Row cache (avoid sstable lookup)
  • Key cache (avoid index scan)

Note: make sure you check the slide for a visual description of the Cassandra read operation. And you can also read the Cassandra Reads performance explained for more details.

Future Direction

• Range delete (delete these cols from those keys)
• Vector clocks (including server-side conflict resolution)
• Altering keyspace/column family definitions on a live cluster
• Byte[] keys
• Compression
• Multi-tenant support
• Less memory restrictions