Over the past few several months a variety of database vendors have announced product upgrades that use silicon: SSD/Flash devices or DRAM, to reduce reliance on disk I/O. As you would imagine, SAP is not surprised by this trend. Our SAP HANA product was built from the ground-up assuming that hot data needed to be in-memory and that more data was becoming hotter all of the time.
We’ve always anticipated vendors would apply the 80/20 rule to extend their disk-based product architectures… but we also know that the 20/80 corollary holds… that it takes 80% of the effort to achieve the last 20% advantage. SAP HANA retains this advantage on the I/O optimization front… and by itself a 20% reduction in the overhead of I/O yields a significant overall performance advantage.
But the HANA database advantage extends far beyond just I/O optimization. HANA is designed using high performance computing (HPC) techniques. A deep commitment to column store provides the ability to process queries using specialized instruction sets: vector processing and SIMD instructions; and to stream data to the processor in ways that fully utilize every core in a modern multi-core CPU. This fundamentally new database architecture processes queries using memory as memory… not as a block I/O device. It is based on a new columnar database engine… not on a row-based engine that uses columns only to reduce I/O. This supercomputing architecture provides another order of magnitude advantage. HANA is a platform that moves application processing to the data rather than move the data over networks to the application. HANA has developed new capabilities including the ability to execute OLTP and analytics against the same schema and at Sapphire next week we will announce new capabilities that will enable HANA to be the foundation for a virtual data warehouse.
Taken together, these features provide the basis for a new business intelligence architecture that does away with most of the artifacts required to support redundant data in separate databases for separate functions. We are already demonstrating that a simple architecture without pre-aggregated tables, indices, or materialized views is possible. With the introduction of the Business Suite on HANA we are demonstrating that a simple architecture where reporting and analytics execute directly against operational data is possible.
It is possible to glue a column table type onto a row-based architecture and provide significant gains. It is possible to redirect block disk I/O to SSD devices and to memory. It is possible to extend the architecture developed to support a multitude of small single-core servers in a distributed environment to accomodate the large multi-core servers that will drive the next generation of computing. It is possible to apply the 80/20 rule again and again. But the 20/80 rule says that there is an opportunity to gain 20% again and again. Capturing and accumulating these gains is what makes HANA special. And these accumulated gains mean that while others will now and again make an 80% gain in one dimension, and trumpet that gain in the market, they will not easily match the overall performance offered by HANA.
VN:F [1.9.22_1171]HANA is More Than In-Memory,