vSphere 8 for SAP NetWeaver and AnyDB is supported and certified since March this year. In August, Lenovo, Pure, Intel, SAP, and VMware have successfully finished the vSphere 8 validation for SAP HANA on 4th Gen Intel® Xeon® Scalable Processor 2-socket based server systems, codenamed Sapphire Rapids.
I am happy to announce SAP HANA support for this combination of technologies.
The performance gains of this platform are significant and the best results for a 2-socket server we have seen with VMware vSphere so far. See SAP support note 3372365 for details on the supported versions and configurations.
SAP HANA on vSphere validation environment and partners
For the vSphere 8 SAP HANA 2.0 validation we used a 2-socket Lenovo ThinkAgile VX650 V3 Certified Node server with Intel® Xeon® Platinum 8490H processors of an external FC based Pure FlashArray/X50 storage system. The addition of an external Pure FlashArray/X50 storage unit has made, not only our validation tests, but also the BWH Benchmark, a very comparable setup to customer landscapes and previous CPU generations tested by the VMware team.
Previously, I’ve provided information on how the platform performed when virtualized with VMware. This time, I will also provide comparisons to older Intel CPU generations,soon to be supported for SAP HANA on vSphere 8. The data provided will help you understand what to expect from the new 4th Gen Intel Xeon scalable processor platform, virtualized with vSphere 8.
SAP HANA on vSphere validations
I receive many questions from customers on why VMware SAP HANA validations take so long.
One reason is we must use production level HW vs prototype HW for our validation tests, eliminating the possibility of an early start since we must wait until the servers are released and shipped to us.
Another reason, especially with new CPU and platform generations: we must find the firmware/BIOS and VMware settings to run SAP HANA optimally with minimal deviations to a physical deployed SAP HANA system. Optimizing VMware vSphere for SAP HANA is important since a hypervisor solution is normally optimized to run multiple VMs on a server system with a fair-trade resource distribution algorithm.
By defining the best configuration and meeting the demanding SAP HANA KPIs, SAP can provide support for SAP HANA on vSphere, and you as a customer can be ensured that SAP HANA performs well and runs reliably on our solution. All this takes time and we do not want to jeopardize the performance or stability of the solution by rushing these validations for the sake of a quick sales cycle.
SAP HANA 4th Gen Intel Xeon (Sapphire Rapids) 2-socket BW performance
The SAP BW Edition for SAP HANA Standard Application Benchmark (BWH) is currently the only available public SAP HANA benchmark. The BWH benchmark has three phases with each focusing on different areas of a system:
Phase 1 – Data Load Phase:
This phase provides information on how quickly data can be loaded from disk to the server memory and is a combination of CPU and IO intensive loads.
Phase 2 – Throughput Phase:
This phase provides information on how well data can be processed by the CPU, by heavily utilizing memory.
Phase 3 – Query Measurement Phase:
This phase measures how long a complex query runs, showing mainly the single thread performance capabilities of the platform.
While comparing different BWH benchmark results is possible, only phases two and three are directly comparable since phase one results depend strongly on the used storage configuration, which is normally different from vendor to vendor. Please note that while most tests were executed during our validation testing, not all tests were officially published, but all test results were verified and checked for accuracy. Certified and published SAP BWH benchmarks can be found here.
The first two comparisons compare internal running tests with a bare metal system running SAP HANA BWH to virtualized systems running BWH. Please note these are internal tests, which are not published benchmarks. Table 1 shows a L-Class configuration, which follows the SAP defined memory configuration for BWH. The configuration is 1.5 TB per NUMA node (3 TB in total).
Table 1. 3 TB BWH BM vs. vSphere 8 SAP HANA system
Table 2 shows a 2 TB per NUMA node, which is a SAP HANA Tailored Data Center Integration (TDI) configuration (4 TB in total) BWH benchmark.
The differences between the physical and virtual deployment are quite small. The published vSphere 8 benchmark (cert. 2023030) can be found here. Details on the Lenovo system used can be found here, and details on the Pure storage system used can be found here.
Please note, the bare metal and virtual benchmarks were performed on the exact same server systems from Lenovo (Lenovo ThinkAgile VX650 V3 CN) with Intel® Xeon® Platinum 8490H processors and benchmark environment, including the storage configuration, which was an external FC based Pure FlashArray/X50 system.
For our soon to be finished 3rd Gen Intel® Xeon® processor vSphere 8 validation we’ve re-run BWH on a 4 TB Intel Xeon Platinum 8380 processor CPU 2-socket system and used the same environment, including storage and network, and the exact same SAP HANA, OS and test tool versions. Table 3 demonstrates how the 4th Gen Intel Xeon CPU had 50% more compute threads but had 66% more queries (QPH) in phase 2 by maintaining the single thread complex query runtime in phase 3.
Table 3. VMware internal vSphere 8 3rd Gen Intel Xeon processor 2-socket vs. 4th Gen Intel Xeon processor 2-socket BWH performance
This demonstrates that the latest 4th Gen Intel Xeon CPUs not only have more threads, but also work more efficiently than the previous generation. The graphic below shows the 4th Gen Intel Xeon processor benefit in query throughput.
VMware vSphere SAP HANA with 4th Gen Intel Xeon processors – 2-socket SAP HANA Mixed-Workload performance
The performance gains are even more profound when comparing 4th Gen Intel Xeon processor results with our internal SAP HANA mixed workload tests. These tests focus on OLAP and OLTP mixed-workload performance, clearly demonstrating that the platform has enough power for the most demanding SAP HANA workloads.
Customers can either go beyond the SAP standard defined memory sizes or, select smaller 4th Gen Intel Xeon CPUs, like 32 or 48 core variants. Doing so requires an SAP workload-based sizing. If a sizing is not possible you can simply select the SAP HANA defined standard configuration sizes, which are based on the 60-core 4th Gen Intel Xeon processor CPU version with 1.5 TB per CPU socket for OLAP (BW) and 2.0 TB per CPU socket for OLTP (ERP) type workloads.
Table 4 provides an overview of recently conducted tests with the same SAP HANA workload with different CPU generations. The Intel E7-8880v4 tests were performed with vSphere 7, but the differences between vSphere 7 and 8 with this CPU are negligible, since vSphere 7 and 8 have the same feature set on an Intel E7-8880v4.
Table 4 – Internal VMware 2-Socket 4th Gen Intel Xeon, vSphere 8 VM vs. vSphere 7 Intel E7-8880v4, vSphere 8 on 2nd Gen Intel® Xeon® and 3rd Gen Intel Xeon Mixed SAP HANA Workload Test VM @ Max-Out CPU utilisation.
The graphic below shows an over 2x performance increase over different CPU generations. The comparison starts with a previous generation 4-socket Intel E7-8880v4 system, and includes an 8-socket 2nd Gen Intel Xeon 8280L system, clearly demonstrating a 2-socket 4th Gen Intel Xeon processor 240 thread CPU system provides higher transactional and query performance than an 8-socket system with 448 physical CPU threads, allowing a significantly reduced number of hosts, due to higher CPU performance and consolidation possibilities, lowering the overall costs related to operating the environment.
The graphic below shows the performance gains with 4th Gen Intel Xeon processor when comparing the different platforms running the same SAP workload.
Please note that the 8-socket 2nd Gen Intel Xeon systems are SAP HANA vSphere supported with up to 12 TB, while with a 2-socket 4th Gen Intel Xeon the SAP HANA default memory size for OLTP (S4/HANA) is 4 TB. In short, there is still a need for larger 4 and 8-socket systems.
Conclusion
The SAP HANA performance gains of the 4th Gen Intel Xeon scalable processors used in the tested 2-socket servers are impressive. Customers still using previous Intel CPU generations should strongly consider this platform for their next system upgrade. Customers concerned with aging server HW can easily upgrade their virtual SAP HANA VM running on Broadwell systems with 4th Gen Intel Xeon-powered systems by simply migrating these VMs over to the new environment and critically aligning the VM configuration to the new platform capabilities. SAP HANA or OS upgrades are not required. A server HW upgrade can be done with minimal downtime -minutes vs hours – when migrating a VMware vSphere virtual deployed SAP HANA system to new server hardware.
What’s next?
We are currently finishing the vSphere 8 validations for 2nd Gen Intel Xeon and 3rd Gen Intel Xeon Scalable platforms. This will allow you to continue to operate these systems with vSphere 8, and soon, we are also planning to release SAP HANA support for 4-socket 4th Gen Intel Xeon systems, after a successful validation with SAP.
8-socket 4th Gen Intel Xeon processor systems are planned for next year. For details about our planned validation and test roadmap please review the following SAP help page.
Please be aware that SAP has migrated the old SAP WIKI to the SAP help portal. All SAP HANA vSphere relevant content can now be found here.
