We propose two alternative solutions to solve the concurrency problems
First compensation-based solution is shown in Figure 1. Here, VM (View
Maintenance), VS (View Synchronization) and VA (View Adaptation) are general
view algorithms based on different types of source updates [12,13,14]. To
solve the concurrency problems, we propose to apply
in the "Query Engine" to handle concurrent data updates and concurrent
rename schema changes. To resolve the concurrency caused by concurrent drop
schema changes , we introduce a new algorithm called "Dyno" to dynamically
schedule the maintenance order of the updates to solve the conflicts. The details
could be found in .
Second multiversion-based solution is shown in Figure 2. Here, A new transaction
model (DWMS_Transaction) is introduced which integrates both base update transaction
and maintenance transactions. We then rephase the DW maintenance anomaly problems
to the serializability of the DWMS_Transactions. To achieve such serializability,
we introduce a multiversion algorithm for concurrency handling, in particular, the
versioned data are stored in the "TxnWrap" . We also explore several optimizations
based on this approach, such as wrapper optimization, parallel DW maintenance, batch
DW maintenance etc [1,4].
We implemented a prototype system on top of EVE system. We also
experimentally evaluated our proposed techniques.