Oracle Database In-Memory…You Rock!

Oracle recently compared SAP’s BW-EML benchmark running on HANA with Oracle’s 12c In-Memory option running on Oracle Exadata. In this test, Oracle used an equal number of CPUs as used in SAP benchmarking and Oracle came in 2X faster!

Generally, SAP does not publish results for any of their standard OLTP benchmarks for their HANA database, so it is difficult to compare HANA’s performance with other databases. But SAP has created a new benchmark called SAP Business Warehouse Enhanced Mixed Workload (BW-EML). This benchmark measures the performance of a database running the SAP BW analytical application.

Oracle ran the initial benchmark with one billion records on one of the database servers in the Exadata X5-2 with 36 cores. Oracle Database 12c was able to complete over 300K navigation steps (NS) in one hour where SAP’s benchmark achieved 148K NS with 36 cores. SAP achieved 1.9M NS/hour with 504 cores where Oracle achieved 2.3M NS/hour with 288 cores. The scale-up test was also 2X faster.

Oracle repeated the same benchmark using two database servers, then four and finally all eight database servers on the Exadata X5-2. Oracle Database 12c was able to show close to perfect scalability as the number of database servers was increased.

Oracle 12c IM running on Exadata has the following features to back the performance of any analytical/OLTP workload:

  • Access only the columns that are needed
  • Scan and filter data in a compressed format
  • Prune out any unnecessary data within each column
  • Use SIMD to apply filter predicates
  • Hybrid compression
  • Smart Flash Cache
  • Storage Indexes

Based on these results, SAP issued several statements. Their claims and Oracle’s response is included below.

SAP Statement Oracle Rebuttal
Oracle manipulated the benchmark of an SAP analytics application running on Oracle Database 12c In-Memory by using a “custom setup” with database triggers and materialized views that aren’t supported in real-world production environments. Both features, which have been included in Oracle Database for years, are used regularly by customers. There’s nothing “custom” about them.
Oracle’s “custom” trigger-and-materialized-views setup can lead to hard-to-spot data inconsistencies. A large percentage of Oracle’s customers take advantage of triggers in a variety of different systems without introducing any data inconsistency.
Oracle misrepresented the true number of processors it used in the benchmark by running the benchmark on its Oracle Exadata Database Machine, which incorporates storage servers that can be used for database processing. All data used in Oracle’s benchmark was processed in memory, in the compute servers. No data was processed in the storage servers. None.
Oracle Exadata Database Machine is much more expensive that a commodity Dell or Lenovo server, which are often used to run databases. The performance advantages provided by Oracle Exadata Database Machine – optimized hardware/software systems that are preconfigured with compute, storage, and networking hardware – more than make up for the difference.
SAP has more than 7,000 customers running SAP applications on HANA, while Oracle has zero customers running SAP applications on Oracle Database In-Memory. Oracle Database 12c In-Memory is, in fact, one of Oracle’s most popular technology introductions. Several prominent Oracle Database In-Memory customers spoke at Oracle OpenWorld 2015; one whose SAP application runs on Oracle Database In-Memory was the subject of a presentation. While many SAP customers have bought HANA, it’s unclear how many are actually using it, as SAP throws in HANA as part of the deep-discount programs for its enterprise applications.

One additional note. Last year I showed Oracle IM to one of my SAP customers. It was before Oracle IM certification by SAP. We used a midsize server to show Database In-Memory in action. At the same time this customer was also evaluating HANA. Currently this customer is planning to go for supercluster.


*** The opinions expressed here are solely mine and do not represent any organization.

« Back to blog