Conservative Consistency
Consistency of replicated data can be kept by using
one-copy serializable
(conservative) consistency schemes.
Conservative schemes ensure that clients access the latest data and data
copies see the same order of updates.
Since all copies see the same order of updates,
users do not have to worry about which copy of the data they are accessing.
Replication becomes transparent for the user.
Thus these schemes provide a good user replication model.
Consistency Strategies
Simple strategies for providing pessimistic consistency include
primary copy, tokens and voting.
Primary copy and token mechanisms decide on a copy from which all data reads
must be done.
Data writes are sent to all other copies.
Voting is a more general technique where a certain minimum number of tokens
are required for data reads and writes.
A write must obtain tokens from more than half the number of copies.
Problem
These schemes prevent inconsistencies by limiting availability.
They assume that each partition could make updates and
thus restrict updates to only one partition.
For several large scale applications like banking, reservation systems,
personal applications like appointment calendars, design documents,
meeting notes and mobile file accesses, continued access is critical even
when network partitions occur.