In my previous post, I introduced IBM DB2 Analytics Accelerator (Accelerator) and explained how it is capable of serving online analytical processing (OLAP) and online transaction processing (OLTP) queries at the same time. Now it is time to go into more detail and explain how it all works.
The DB2 optimizer is the key
The design concept is quite simple. The DB2 for z/OS optimizer is aware of DB2 Analytics Accelerator’s existence in a given environment and can execute a given query either on the DB2 Analytics Accelerator or by using the already well-known access paths within DB2 for z/OS. The DB2 optimizer decides which queries to direct to the Accelerator for hardware-accelerated parallel query processing, thus the Accelerator is essentially transparent to the applications and reporting tools querying DB2 for z/OS.
So, the DB2 optimizer determines whether a query is best suited to run utilizing symmetric multiprocessing (SMP), leveraging DB2 for z/OS; or the hardware-accelerated massively parallel processing (MPP) architecture delivered by the PureData System for Analytics, powered by Netezza technology. It chooses the best option transparently. There are essentially no changes to the way you query the DB2 system.
After defining tables to be accelerated and then loading data into the Accelerator, you essentially have an environment that provides both an SMP and MPP “personality” to the same table. This allows you to leverage the legendary DB2 for z/OS qualities of service and performance for transactional queries, as well as Netezza performance for complex queries.
Complex queries are automatically routed by DB2 for processing by the Asymmetric Massively Parallel Processing (AMPP) engine, an architecture that uses special-purpose hardware accelerators (a multi-engine field-programmable gate array [FPGA]). These accelerators decompress and filter data for relevance to the query before it is loaded into memory and given to the processor for any necessary aggregation and final processing. This entire process is transparent to the application and user. The user only notices an improvement in the speed at which the query is resolved by the system, specifically the analytical ones. The system in general also benefits, because the analytical load is transferred out of DB2, and thus there is more capacity available to serve additional transactional queries. And keep this in mind: once a query is derived to be solved by the Accelerator, its execution does not consume MIPS.
High Performance Storage Saver
Once you have this system installed, you may use it for more than accelerating analytical queries. There are some interesting use cases that will help you obtain more benefits from your hybrid system, such as the High Performance Storage Saver.
In this use case, you can save in Total Cost of Ownership (TCO) by using the Accelerator to store historical data to reduce storage costs on z Systems. This way, you are not consuming storage in your IBM z Systems. Instead, you are using the lower cost of storage in the DB2 Analytics Accelerator, while still being able to access that historical data and perform analysis on it. And this is the most interesting point . . . the historical data that you would typically take out of your mainframe due to the cost of storage in that system can now be explored online. You do not need to recover the tapes where you store the historical data to access it.
Simple is still better
Meanwhile, I encourage you to learn more about DB2 Analytics Accelerator, where you are leveraging the “simple is still better” design tenents of PureData System for Analytics And of course, I am more than happy to read about your opinions and see you share your experiences in the comments area below or join me in a conversation on Twitter.
See additional posts
Isaac is working as a data warehouse and Big Data technical pre-sales professional for IBM, covering customers in Spain and Portugal, where his special focus is on PureData for Analytics. He joined IBM in 2011, through the Netezza acquisition. Before that, he has held several positions in pre-sales and professional services in companies such as Oracle, Sun Microsystems, Netezza and other Spanish companies. During the years previous to working at IBM, he has acquired a diverse experience with different software tools (databases, identity management products, geographical information systems, manufacturing systems…) in a very diverse set of projects. He also holds a Master of Science Degree in Computer Science.