The arrival of the Lunar New Year brings new fortune and new opportunities. If you have been following my blog posts for the last 2 years, it has been quite a journey enabling Optane with our OEM's, GSI and Cloud Service Providers. Here is our latest and greatest news relating to Intel Optane persistent memory (PMem) with SAP HANA. As a quick refresher: Intel Optane PMem encrypts your data stored in the HANA Database, increases memory density which allows lower TCO as a result and customers can run Translytic Workloads (Eg. S4HANA Group Reporting) to simplify their architecture.
In the past, I outlined how to configure Intel Optane PMem with SAP HANA and how to seamlessly test a production ECC On HANA workload & Sizing SAP Workloads with 2nd Generation Intel® Xeon® Scalable processors using Intel Optane PMem 100 series.
Today, we’re unveiling new performance metrics for Intel Optane PMem 200 series that bring greater value to customers’ applications and workloads.
Flexibility meeting customers challenges
Today we have customers running Intel Optane PMem in production environments using 4-socket, 6-socket, and 8-socket server workloads for ECC on HANA, S/4HANA, BW On HANA & SCM on HANA.
We’re dedicated to walking the talk: our customers validated our memory density claims when they converted their scale-out workloads to scale-up workloads for SAP Business Warehouse on SAP HANA. As a result, one such customer simplified and consolidated their architecture from 21 scale out systems to 3 scale up systems with increased memory density.
Subsequently, our team collaborated with the Microsoft Azure team to show how using Intel Optane PMem (such as S224OM, S224OOM, S448OM, and S896OM instances) with DRAM led to a 30 percent cost savings for systems using Intel compared to instances using DRAM alone.
Today, we also have multiple Azure customers running SAP S/4HANA scale-out and scale-up solutions with Intel Optane PMem using 8-CPU and 16-CPU designs. It’s a thrilling achievement, and we’re excited for their “go lives” this spring particularly for the world’s largest S/4HANA database of 48 TB (as of this writing).
Comparing Intel Optane PMem 100 series & PMem 200 series with SAP HANA BWH benchmark
Since we introduced Intel Optane PMem 100 series, multiple Original Equipment Manufacturers (OEM) have submitted SAP BWH benchmarks for 2, 4, 8, and 16-CPU designs for SAP HANA Tailored Datacenter Integration (TDI) configurations, which allow SAP HANA customers to make use of existing hardware and infrastructure components.
The first benchmark done by Microsoft compared DRAM and Intel Optane PMem 100 Series. Microsoft Azure PMem instance S896OM( 12 TB DRAM + 12 TB PMem) results are 99.4% compared to Azure DRAM instance S896M(24 TB DRAM).
The two SAP Benchmarks (SAP Certification : 2021028 using SAP HANA 2.0 Benchmark Version 3 and SAP Netweaver 7.50, 2021003 using SAP HANA 2.0 Benchmark Version 3 and SAP Netweaver 7.50) we had compared are like for like since they have an overall combined capacity of 24 TB HANA Database.
Second Benchmark done by IBM Cloud and Lenovo compares Intel Optane PMem 100 Series using 2nd Generation Xeon Scalable Processors & PMem 200 Series using 3rd Generation Xeon Scalable Processors.
The two SAP Benchmarks(SAP Certification : 2021053 using SAP HANA 2.0 Benchmark Version 3 and SAP Netweaver 7.50, 2021009 using SAP HANA 2.0 Benchmark Version 3 and SAP Netweaver 7.50) we had compared are like for like because both of them loaded 5.2 Billion rows using the SAP BWH benchmark and they both support 3 TB HANA Database.
Here is how to interpret the results of SAP BWH Benchmark
An SAP BWH Benchmark consists of 3 Phases :
- Phase 1 (data load). Phase 1 consists of loading a certain number of records to learn how the system performs. Lower numbers equal better performance. Every customer using SAP HANA Data Warehouse or SAP Business Warehouse runs these jobs nightly. In short: the quicker the jobs complete, the more time the system is available for users to run reports. In the above illustrated benchmark, approximately 5.2 billion records were loaded in 12,859 seconds versus 32,856 with previous generation
- Phase 2 (queries executed per hour). This denotes the number of queries a system can run within an hour. The higher the number, the better the performance. Customers size SAP systems for peak performances so it can handle critical month end tasks like Finance Month end close or Inventory runs tasks efficiently. Intel Optane PMEM 200 Series can run 4222 concurrent queries within an hour vs. 1911 with previous generation as per the second benchmark
- Phase 3 (single large complex query). This measures the time it takes to run a complex query, based on a real-world use case. The lower the number the better the performance, the complex query took 89 seconds vs. 125 seconds with previous generation as per the Second Benchmark
What contributed to the higher performance?
1. Each 3rd Generation Intel Xeon Scalable processor supports 8 Memory channels(a total of 8 DRAM DIMMS + 8 PMem DIMMS) vs. 6 memory channels (a total of 6 DRAM DIMMS + 6 PMem DIMMS) with 2nd Generation Intel Xeon Scalable processors
2. More cores & powerful cores with 3rd Generation Intel Xeon Scalable processors
3. 32% more Bandwidth in our Intel Optane PMem 200 Series
Let's talk TCO
Prior to this role I was with IT managing our Internal Intel SAP HANA systems. One of my favorite metric to measure performance/price was Cost Per Terabyte for SAP HANA systems. Let's say you have a 5 system Landscape(Development / Quality Test / Production/ High Availability & Disaster Recovery) for your S4HANA or B4HANA systems. Your Bill of Material from your OEM Vendors or Cloud Service Providers would be the cost of Database Servers(can be DRAM or Intel Optane Persistent Memory) which is what your compute costs would be.
Cost Per Terabyte = Total BOM Cost of Database Servers / Total Memory in terabytes in all the Database Servers
With some help from team Lenovo I used the Lenovo Datacenter Solution Configurator and priced out 3 configurations
Config 1 : PMem 100 Series 3 TB HANA DB with 2nd Gen Xeon Scalable Processor Cost Per Terabyte is $47,600
Config 2 : PMem 200 Series 6 TB HANA Database 3rd Gen Xeon Scalable Processor Cost Per Terabyte is $32,000
Config 3 : DRAM Based 4 TB HANA Database with 3rd Gen Xeon Scalable Processor Cost Per Terabyte is $57,300
Intel Optane PMem 200 Series (Config 2 above) has 34% lower Cost Per Terabyte compared to an all DRAM system(Config 3)
Conclusion: The benefits are measurable
In my role I get to talk to multiple customers in a technical capacity answering various questions about how to size, deploy and architect SAP HANA Solutions with Intel Optane PMem.
Here are the Top 3 reasons customers say they chose Intel Optane Persistent Memory:
- Lower TCO
- Protects against unplanned downtimes
- Simpler architectures with increased memory density
TCO for 1st and 2nd Generation Intel Optane persistent memory
- 30% lower monthly costs with Microsoft Azure HANA Large Instances using Intel Optane PMem with 2nd Generation Intel Xeon Scalable Processors
- 34% Lower Cost Per Terabyte for SAP HANA for On-Prem Systems using 3rd Generation Intel Xeon Scalable Processors
Performance comparison between 1st and 2nd generation
- 99.4% for Phase 1 and Phase 2 SAP BWH Benchmarks compared to all DRAM system using Intel Optane PMEM 100 Series
- Double the performance for Phase 1 and Phase 2 SAP BWH benchmarks compared to 1st generation Intel Optane PMem 200 series
- 32% more bandwidth with PMem 200 series and powerful 3rd Generation Xeon Scalable processors leads to faster Phase 2
Increased Memory Density
- 6 TB of SAP HANA database using 2s PMem 200 series and powerful 3rd Generation Intel Xeon Scalable processors
Krishna Yalamanchi is a Technical Solutions Specialist at Intel in the Memory Solutions Product Group. He focuses on enabling Intel Optane persistent memory solutions with Hyperscalers, Global System Integrators (GSIs) and OEMs.