Extending Database Accelerators for Data Transformations and Predictive Analytics

The IBM DB2 Analytics Accelerator (IDAA) integrates the strong OLTP capabilities of DB2 for z/OS with very fast processing of OLAP workloads using Netezza technology. The accelerator is attached to DB2 as analytical processing resource – completely transparent for user applications. But all data modifications must be carried out by DB2 and are replicated to the accelerator internally. However, this behavior is not optimized for ELT processing and predictive analytics or data mining workloads where multi-staged data transformations are involved. We present our work for extending IDAA with accelerator-only tables, which enable direct data transformations without any necessary interventions by DB2. Further, we present a framework for executing arbitrary in-database analytics operations on the accelerator while ensuring data governance aspects like privilege management on DB2 and allowing to ingest data from any other source directly to the accelerator to enrich analytics e. g., with social media data. The evolutionary framework design maintains compatibility with existing infrastructure and applications, a must-have for the majority of customers, while allowing complex analytics beyond read-only reporting.