Consistency is an important property for systems to have to keep a state that makes sense.

Consistency means that all nodes return the same value (i.e. all nodes have the same view of the data).

• e.g. in this case, users on the west and east coast of Canada receive the same result from the same query

Availability means that node failures do not prevent operation of the system.

• e.g. in this case, even if a server goes down, users in a region where a server is still running can still be active

Partition Tolerance means that the system operates in the presence of a network partition.

• e.g. Even if link goes down, customers on west/east coast are unaffected, aside from being unable to connect to one another

CAP Theorem

There are 3 properties of a distributed system:

• Consistency - all nodes see the same values at the same time
• Availability - If a node fails, the network continues to operate
• Partition Tolerance - the overall system operates despite network partitions

A version of the theorem states that all 3 of the above properties are not possible - only 2 are possible.

This suggests there are 3 kinds of distributed systems: