It may have gotten lost in the amount of HANA-related announcements that were made at the SapphireNOW conference in Orlando this week, but it is a significant achievement that we have not only a new SAP NetWeaver BW benchmark, but new results to report as well.
Realizing how fast analytical requirements change over time, the SAP Benchmark group embarked on an effort to create a new BW benchmark, called the Enhanced Mixed Load benchmark. The goals behind this benchmark were to take the existing BW Mixed Load benchmark, and update it with:
- Query complexity: the EML benchmark uses randomized navigation paths and changing displayed characteristics which are more reflective of ad-hoc queries common in customer environments today.
- Nearer to real-time data: With 10 delta loads running concurrently every 5 minutes, a total of 120 loads per hour, loading is almost constant while user queries are being executed.
- More detailed data. DataStore Objects (DSOs) have become a critical component in customer data models. As such, with this new benchmark, queries run across 10 InfoProviders, of which only 3 are structured InfoCubes, and the other 7 are DSOs containing detailed transactional data. Neither the prior BW benchmark, or industry standard benchmarks like TPC, use detailed transaction data.
In discussions with customers, there was much interest in such a new benchmark, which would be more reflective of how SAP customers are using BW today.
I had the pleasure of working with HP, which was the first SAP partner to volunteer to fully test this new benchmark. The platform they chose was the SAP HANA database running on the HP AppSystem for SAP HANA. Using a “Medium” size appliance with 4 CPUs and 512GB of RAM for the database server, the BW-EML benchmark was loaded with 1 billion rows of data to begin the test, and added another 1 million rows over the course of 120 delta loads.
The results of the benchmark are an impressive 65,990 ad-hoc navigation steps per hour. This equates to roughly 1,000 active BW users (assuming 60 second analyst think time between queries) and 120 concurrent delta loads throughout the hour. What makes this result even more impressive, and is attributable to the capabilities of the SAP HANA platform, is that the BW configuration was essentially running “out of the box,” as highlighted by the following points:
- Caching was turned off in the BW application layer.
- No aggregates were created or maintained during the testing, nor any secondary indexes, pre-calculations or caching of any kind.
- Cube structures were simplified, as is standard with BW running on HANA, by using only a single fact table which supports simultaneous querying and loading, as well as removing the DIMID tables. Multiple fact tables and DIMID tables were implemented in versions of BW to address performance issues with disk-based databases that no longer apply in the case of HANA.
Many thanks go out to the hard work of the HP and SAP benchmarking teams to pull off a great new benchmark, and superb results.