Analysis of data replication collision handling strategies

Analysis of data replication collision handling strategies


Abstract:-

A replication system cannot be reliable and effective unless it can prevent, detect and flexibly resolve collisions, in that order. This seminar evaluates collision handling strategies and compares their strengths and weaknesses. At minimum, a reliable collision strategy must:

• Efficiently prevent all key collisions.
Key collisions are typically impossible to resolve automatically. Globally unique IDs (GUIDs) prevent key collisions outright, so that resolution is never needed. Not all products support GUIDs, most of those that do use large and inefficient GUIDs.

• Prevent all false update collisions.
Updates at different sites to different parts of the same record should often be allowed. Incorrectly flagging them as collisions imposes a heavy administrative burden and greatly limits scalability. False positives account for the majority of update “collisions” in replication products.

• Always detect true update collisions.
Some collision strategies avoid false collisions by failing to detect collisions at all. This is a serious integrity problem.

• Provide flexible collision resolution.
To resolve collisions correctly, a handler must be able to inspect the full data values, the sources (including the originating user and site), and the absolute update times (independent of time zones and relative clock drifts) of all competing updates.
Strategies, which do not cover all four essentials, are not suitable for enterprise data distribution.
Thus we will see all four and concentrates on collision prevention, not just collision repair.