Near real-time analytics with IBM DB2 analytics accelerator

Daniel Martin, Oliver Koeth, Johannes Kern, Iliyana Ivanova
2013 Proceedings of the 16th International Conference on Extending Database Technology - EDBT '13  
The IBM DB2 Analytics Accelerator (IDAA) implements the vision of a universal relational DBMS that processes OLTP and analytical-type queries in a single system, but on two fundamentally different query engines. Based on heuristics in DB2 for z/OS, the DB2 optimizer decides if a query should be executed by "mainline" DB2 or if it is beneficial to offload it to the attached IBM DB2 Analytics Accelerator that operates on copies of the DB2 tables. In this paper, we introduce the "incremental
more » ... " functionality of IDAA that keeps these copy tables in sync by employing replication technology that monitors the DB2 transaction log and asynchronously applies the changes to IDAA. This enables near real-time analytics over online data, effectively marrying traditionally separated OLTP and data warehouse environments. With IDAA, analytic queries can access data that is constantly refreshed in contrast to traditional warehouses that are updated on a daily or even weekly basis. Without any changes to the applications and without the need to introduce cross-system ETL flows, an existing operational data store can be used for data warehousing as well. The analytic query performance provided by IDAA makes it possible to execute reports directly against the transactional schema, thus avoiding the need for costly design and maintenance of a separate reporting schema. Additionally, the Accelerator shields DB2 for z/OS as the transactional system from performance degradation caused by the analyt-...$15.00. ical workload and the replication component synchronizes all data changes in near real-time. We present the architecture of the integrated replication component of IDAA and discuss design decisions that we made when combining the different technologies as well as performance characteristics of the resulting system.
doi:10.1145/2452376.2452444 dblp:conf/edbt/MartinKKI13 fatcat:jlrtnnsx6jhtlpfso2bs5oe67a