Data fragmentation

Data fragmentation

1. Correct if we guarantee

   • Completeness
     • Given a relation R and a set of fragments, any data item (either a tuple or a set of attributes) of R can be found in at least one fragment. Thus, data are not lost when fragmenting
   • Disjointness
     • Given a relation R and a set of fragments, any data item placed in a fragment cannot be found in any of the other fragments. Re- member that data replication must be considered a posteriori, in the allo- cation stage, as a task independent of data fragmentation
   • Reconstruction
     • Given a relation R and a set of fragments, the original re- lation can always be reconstructed from the fragments by means of rela- tional algebraic operators

2. Horizontal fragmentation

   • A selection predicate is used to create different fragments and, according to an attribute value, place each tuple in the corresponding fragment.
   • Example
     • We want to geographically distribute data about projects
       • The predicates to be used would be city = NAME_CITY (where NAME_CITY is each of the city attribute values: Barcelona, Hong Kong, Portland, Glasgow or Frankfurt). Each predicate will result in a fragment
   • When
     • When it needs to mirror geographically distributed data (each node main- ly accesses data related to itself)
       • Facilitate recovery and parallelism
       • Reduce the depth of indexes (since each fragment has its own index, the number of indexes increases but their size is reduced
       • Reduce contention
         • Contention occurs when we are denied access to a database resource because of conflicts with other users, normally related to concurrency control techniques
         • Since several mutually exclusive users, who work on different nodes, can access different fragments simultaneously and no conflicts will be caused
   • What extend should we fragment?

3. Vertical fragmentation

   • Example
     • We want to store the name , city and country attributes at each node, whereas the other data will be stored at the Hong Kong head office.
       • project(pno,name,city,country) and project(pno,budget,category,productivityRatio,income)

4. Hybrid fragmentation

   • Combines both techniques
     • Further fragmenting fragments produced by a previous fragmentation strategy
     • Example
       • We want to geographically distribute data about projects, but only those attributes of interest at each node (the others will be stored in Hong Kong)
         • VF 1 ( ∏ project(pno,name,city,country) ) and ......................................... VF 2 ( ∏ project(pno,budget,category,productivityRatio,income) )
           • VF 2 will be placed at the Hong Kong node,
           • VF 1 will be further fragmented by applying the horizontal fragmentation (i.e., city = NAME_CITY)