Not logged in » Login
Dec 06 2019

Performance Results for SAP HANA with Intel Xeon Optane DC Persistent Memory

Image

Persistent Memory for SAP HANA, powered by Intel Optane DCPMM (Data Center Persistent Memory), is the latest “paradigm shift” innovation for SAP HANA, enabling memory data retention if power to the server is lost or the server reboots. This feature brings tremendous benefits in terms of the uptime of a SAP HANA system and increased service levels. Even large databases of 4 or 6 TB are available within minutes due to the faster start time. When compared to standard DRAM and low-tier storage, DCPMM technology provides far greater capacity for SAP HANA on a single-node server (up to 24 TB as of now) combined with highest-ever performance and lower TCO.

More and more customers are adopting this “next level of SAP HANA technology”, which reflects SAP’s new strategic direction to provide a variety of use cases, and the benefits are clear to see, especially given the rate at which database sizes are growing.  DCPMM presents a very interesting option to consolidate multi-node/scale-out systems to single-node/scale-up configurations e.g. for BW on HANA. Such consolidation offers a remarkable reduction of complexity and maintenance costs as well as bringing down investment costs. This is an important, additional use case, made possible by this Intel-based innovation.

 

 Image

 

In addition to the afore-mentioned benefits, let’s take a closer look at the first test results which analyze the performance SAP HANA customers can obtain when they install systems with DCPMM. Fujitsu and Intel SAP experts analysed two “typical” project scenarios: a 4-socket installation and, just last month, a high-end scenario with 8-socket server systems.

The 4-socket performance test scenario was measured earlier this year, right after the Intel launch. It served as a comparison in terms of the performance for SAP HANA with and without DCPMM, running on identical configurations based on Fujitsu PRIMERGY RX4770 M5. A whitepaper gives interesting insights not only into the results, but also with regard to the basics of Persistent Memory and the use cases for SAP HANA. Find out more here.

Complementary to this, recently-published benchmark scenarios based on “SAP BW edition for SAP HANA Standard Application Benchmark Version 3” (BWH Benchmark) provide very interesting insights regarding performance in high-end SAP HANA installations, comparing the use of DRAM only, “standard” up to now, with the use of Persistent Memory for SAP HANA. Such new memory architecture usually uses a (smaller) part of DRAM and a (larger) part of DCPPM (ratios DRAM:DCPMM of 2:1, 1:1, 1:2 and 1:4 are supported).

The basis for comparison is the “SAP BW edition for SAP HANA Standard Application Benchmark, Version 3.0” (BWH), in this case with 20.8 billion initial records. You can find out more about the benchmark specifications in the footnotes. The SAP Benchmark certificate is available here, while the Generic SAP benchmark explanations can be found here.

The following table gives you the achieved BWH benchmark results:

Image

 

These results can be compared with a recently certified benchmark in the same category, but without DCPMM and only with DRAM. The benchmark was certified on a much larger HPE server** with 16 sockets and 12 TB DRAM, instead of only 9 TB (3TB DRAM, 6TB DCPMM) on the Fujitsu PRIMEQUEST system.

Let us look at the results in more detail, as the benchmark consists in the measurement of 3 different phases: 

  • The Data load phase (Phase1) is a combination of CPU- and IO-intensive load. Despite the higher number of CPUs, 16 instead of 8, they were not able to take advantage of it.
  • The Query Throughput Phase (Phase 2) is CPU bound, which is why the 16-socket system achieved a higher throughput. That is clearly due to the doubled number of CPUs.
  • The Query Runtime Phase (Phase 3) generates a small load. Only a few processor cores are used, and single-thread performance is important for short runtimes. The 16-socket system is only slightly better here, although only DRAM was used, and no DCPMM.

 

The test and benchmark results demonstrate that:

  • Intel® Optane™ DC Persistent Memory (DCPMM) has an excellent performance with no significant performance impact on SAP HANA, compared to DRAM-only systems;
  • an 8-socket system can now achieve scales which were only possible in the past with bigger and much more costly systems;
  • Fujitsu PRIMEQUEST 3800B2 can handle huge amounts of data with a 9 TB HANA memory, proven with the BWH benchmark – and even more up to 24TB RAM for extremely large HANA systems;
  • The 2nd generation of Intel® Xeon® Scalable Processors with DCPMM offers new options for HANA configurations with many advantages, typically around 10x reduced restart times, bigger and affordable memory configurations and the chance to consolidate scale-out configurations which result in reduced maintenance costs.


Fujitsu and Intel SAP experts of Global SAP Competence Center and Benchmark specialists work in close collaboration to drive innovation for SAP HANA, continuing the long track record which stretches back even further than the 20 years of SAP on Linux, for the benefits of our mutual SAP customers.

 

Footnotes:

“Intel, the Intel logo, Intel Inside, the Intel Inside logo, Xeon, and Optane are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other countries.”

SAP Benchmark Source: http://www.sap.com/benchmark

SAP BW edition for SAP HANA Standard Application Benchmark Version 3, 3-tier, Scale-up, 20.8 billion initial records:

* Fujitsu Server PRIMEQUEST 3800B2 , Intel Xeon Platinum 8280L processor, 8 processors, 224 cores, 448 threads, 3072 GB DRAM and 6144 GB Persistent Memory, SUSE Linux Enterprise Server 12, SAP NetWeaver 7.50, SAP HANA 2.0: 19635 seconds runtime of last data set, 3621 query executions/h, 168 seconds total runtime of complex query phase, cert.no. 2019059.

** HPE Superdome Flex, Intel Xeon Platinum 8280L processor, 16 processors, 448 cores, 896 threads, 12288 GB DRAM, SUSE Linux Enterprise Server 12, SAP NetWeaver 7.50, SAP HANA 2.0: 19816 seconds
runtime of last data set, 7158 query executions/h, 161 seconds total runtime of complex query phase, cert.no. 2019050.

 

Rainer Hettinger

 

About the Author:

Rainer Hettinger

Principal Business Developer, SAP Category Management EMEIA

SHARE

Comments on this article

No comments yet.

Please Login to leave a comment.

X

Please login

Please log in with your Fujitsu Partner Account.

Login


» Forgot password

Register now

If you do not have a Fujitsu Partner Account, please register for a new account.

» Register now