Examine critical components of Cloud computing with Intel® software experts
109 Discussions

Scale UP -- and up -- with SAP HANA and Persistent Memory

0 0 688

Through joint innovation between HPE and Intel, a new milestone has been achieved for SAP HANA with persistent memory. Think bigger—but also faster with HPE Superdome Flex and Intel® Optane™ persistent memory technology.

First the milestone. Leveraging the modular, scale-up architecture of the HPE Superdome Flex server, an SAP Business Warehouse (BW) Edition for SAP HANA benchmark new world recordi was set using a system with 16 Intel® Xeon® Platinum 8280L processors, 12TB of DRAM, and 12TB in 128 GB DIMMs of Intel® Optane™ persistent memory technology. Here’s a quick benchmark snapshot.


You can find complete information on the SAP Standard Application Benchmarks website.

At first blush, the benchmark shows you can scale up memory capacity using Intel Optane persistent memory in combination with DRAM for massive SAP HANA environments. Perhaps no surprise given SAP HANA has been optimized for persistent memory. Yet there’s more, as this benchmark also shows very large data warehouses can indeed run faster with more processing power. But wait, isn’t the ability to run larger databases on smaller systems a potential benefit of persistent memory given the larger module capacity vs. DRAM? Yes, as your system size limits allow.

But let’s take a look at other considerations between the two.

Using an all-DRAM system delivers very high performance. DRAM is volatile, however, so if a server needs to be restarted, you must wait for data to load from storage before accessing SAP HANA—which can be over an hour for large databases. Restarts of production systems are uncommon (system upgrade or rare outage) and user disruption can be avoided using a failover high availability (HA) server and SAP HANA replication. For test and development systems, however, restarts are much more common, and the time lost for data to load can be hours. With Intel Optane persistent memory, restart times can shrink to minutes—up to 13x faster than using DRAM aloneii.

Systems used for disaster recovery (DR) can also benefit from persistent memory. Since these systems are seldom put into action, enterprises often make use of them by running other non-critical applications. If a disaster strikes, applications are powered down, SAP HANA is started, and data is read from disk. But if the SAP HANA data can be kept in persistent memory, IT can bring DR systems online sooner. And as users will often accept some drop-in application speed in a disaster situation, gains from less downtime can outweigh differences in performance when compared to an all-DRAM system.

So yes, the additional capacity of Intel Optane persistent memory allows you to run larger databases on systems with fewer processors. Want to run faster? Scale up your processing power as well. As this benchmark shows, HPE Superdome Flex provides the muscle of Intel Optane persistent memory and brawn of Intel® Xeon® Scalable processors for the fast restart of large SAP HANA databases. Plus, with Superdome Flex, you can start small and scale up as your needs increase. You won’t outgrow and face costly system replacement, but you also avoid having to overprovision capacity that you don’t need today.

One last thing. As you scan the list of SAP BW edition for SAP HANA benchmarks, notice something else. Whether systems with all DRAM, or with DRAM and Intel Optane persistent memory, Intel® technology-based systems significantly outperform non-Intel technology-based systems (based on query executions per hour). Just one more reason why over 90 percent of servers powering SAP HANA leverage Intel Xeon processorsiii.

Learn more about HPE Superdome Flex, with scale-up and scale-out capacity for SAP HANAiv.
Learn more about Intel Optane persistent memory technology.
Learn more about Intel® Xeon® processors
Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations, and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more complete information visit

Performance results are based on testing as of the date set forth in the configurations and may not reflect all publicly available security updates. See configuration disclosure for details. No product or component can be absolutely secure. Intel, Xeon, Optane, and the Intel logo are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other countries.

© Intel Corporation

[i]SAP NetWeaver 7.50; Scale-up; SUSE Linux Enterprise Server 12; 6 x Intel Xeon Platinum 8280L processor, 2.7 GHz, 448 cores, 896 threads, Memory (DRAM) 12,288 GB, Persistent Memory 12,288 GB; Phase 1: Data Load - 31,870 sec, Phase 2: Query Executions per Hour – 3,412, Phase 3: Runtime of Complex query phase - 262 sec
[ii]Based on Intel testing as of March 1, 2019. Columnar store entire reload into DRAM for 1.3 TB dataset is 20 mins. Entire system restart before is 32 minutes and with Intel® Optane™ persistent memory is 13.5 minutes (12 mins for OS + 1.5 mins). Configuration details: baseline: 4S Intel® Xeon® Platinum 8280M processor (28 cores), 6 TB memory (48 x 128 GB DDR4 at 2,666 megatransfers per second [MT/s]), 10Gb Intel® Ethernet Converged Network Adapter X520, 60 x 480 GB Intel® SSD DC S4600 Serial ATA (SATA), BIOS: WW48’18, SUSE 15*, Intel® IT workload, 3 TB SAP HANA® database, security mitigations: variants 1, 2, 3 enabled. AD 2-2-2 config: 4S Intel Xeon Platinum 8280L processor (28 cores), 9 TB memory (24 x 256 GB Intel® Optane™ DC persistent memory, 24 x 128 GB DDR4 at 2,666 MT/s), 10Gb Intel Ethernet Converged Network AdapterX520, 90 x 480 GB Intel SSD DC S4600, BIOS: WW48’18, SUSE 15, Intel IT workload, 6 TB SAP HANA database, security mitigations: variants 1, 2, 3 enabled.
Based on Intel testing as of March 1, 2019: base configuration: 10 systems with 4S Intel® Xeon® processor E7-8894 v4, 768 GB (12 x 64 GB) memory. Compared to: 5 systems with 4S Intel Xeon Platinum 8280L (28 cores), 2,304 GB (6 x 256 GB Intel® Optane™ persistent memory + 6 x 128 GB DRAM, 2-2-2, App Direct Mode). Base system included $35,592 on CPU, $33,994 on memory, $24,000 on storage, $7,603 on RBOM, and $0 on software, for a total of $101,189 (or $1,011,891 for 10 systems; $67,459 per TB of storage). Comparison configuration included $71,624 on CPU, $123,163 on memory, $54,000 on storage, $7,603 on RDOM, and $0 on software, for a total of $256,390 (or $1,281,950 for 10 systems; $42,732 per TB storage).
[iii]Gartner Research Note - How to Select Your Optimal SAP HANA Systems Vendor
[iv]SAP Certified and Supported SAP HANA® Hardware Directory