Persistent memory enhancements on IBM Power Systems for SAP HANA

It shouldn’t surprise anyone that the IBM strategy for existing and upcoming memory technologies is anchored in embracing the open ecosystem that gives clients choices to help solve real problems. When IBM and SAP jointly announced support for SAP HANA on IBM Power Systems four years ago, IBM didn’t simply follow what other SAP certified hardware vendors were doing. Instead, we analyzed the problems that clients had and created innovative solutions to help solve them with virtualization and flexibility, resiliency and great performance to top it all off. The market reaction shows thousands of clients embracing SAP HANA on IBM Power Systems, and we’ve experienced phenomenal growth in the Enterprise Linux market.

Our approach is very similar when it comes to persistent memory. Instead of embracing a single persistent memory technology and vendor, we are executing a multi-step approach designed to work with multiple memory technology vendors and have a more comprehensive range of solutions that clients will be able to deploy, in some cases, without even having to purchase new hardware at all.

Client requirements and pain points

Clients have shared with us the pain points they suffer as they attempt to tackle the costs of managing their HANA on x86 deployments. In general, the typical feedback is that there are frequent planned or unplanned outages, which leads to lots of wasted time waiting for systems to shut down, be brought back up, be patched and have all data reloaded.

At IBM, we researched and listened to SAP clients and realized that:

  • Clients demand choice, and all SAP certified x86 vendors who are embracing Intel Optane persistent memory are, in essence, providing the same single vendor proprietary technology to try to fix a problem, that is, frequent hardware outages, that’s prevalent in x86 systems (see the 2019 ITIC report for server reliability survey results).
  • The feedback we received from our HANA on POWER clients was that Power Systems clients do not face the challenge of frequent hardware-related outages. In fact, one large SAP HANA client told us they haven’t had to reboot their Power System for 18 months straight due to any hardware outages. The Forrester TEI Study of IBM Power Systems for SAP HANA determined an average reduction of 48 hours of annual downtime after customers move from non-POWER environment to Power Systems. What POWER clients say they want is a solution that helps to fast restart the environment when there’s a software-related planned (patch the OS or SAP environment) or unplanned outage in their SAP HANA environment.
  • Clients (on x86 or Power Systems) don’t want to trade performance for persistence.
  • Clients were also very clear that they don’t want to add another single point of failure in their infrastructure layer or add a component that will make the solution less flexible by adding more bottlenecks to their infrastructure layer.
  • Clients want to add the capability of persistent memory without having to rip and replace or purchase expensive add-ons to hardware they just bought.

IBM Power Systems vision for persistent memory

With these client pain points in mind, our vision is one that starts with:

  • What is possible today. Clients should evaluate and embrace two new capabilities SAP recently introduced to reduce downtime (Fast Restart option) and reduce the memory footprint by using SAP Native Storage Extension. In both cases, IBM’s technology leadership in I/O performance with PCIe Gen4, memory bandwidth and overall performance leadership mean an enhanced client experience on IBM Power Systems. See SAP Sales and Distribution Benchmark of IBM Power Systems.

Read more

  • Delivering persistence without compromising performance. Later this year (see disclaimer below), IBM intends to enhance our advanced virtualization platform (IBM PowerVM) with a new feature called PowerVM Persistent Memory (PMEM). PMEM is designed to create persistent memory volumes using the existing DRAM technology that our clients already own. By maintaining data persistence across application and partition restarts, it is engineered to allow IBM clients to leverage fast restart of a workload using persistent memory for the vast majority of their planned maintenance and unplanned outages without compromising the performance of HANA during normal use. IBM plans to make this capability available without changes to existing applications.

Read more

  • Embracing an open ecosystem of hybrid memory technologies. Over the next two to three years, IBM believes there will be a broad range of new memory technologies designed to disrupt the industry, providing persistence characteristics with a range of performance and cost that will lower the hardware costs without slowing down in-memory workloads like SAP HANA. These technologies will come from a multitude of memory partners, not a single one, and IBM anticipates that we will be right in the middle of it leveraging industry standards.

Read more

Please visit the links above to go deeper into each of these topics and get a better understanding of our vision and how IBM and SAP intend to execute on that vision to create choice and allow our clients the flexibility they need to unleash the power of SAP HANA.

The information mentioned regarding potential future products is not a commitment, promise, or legal obligation to deliver any material, code or functionality. Information about potential future products may not be incorporated into any contract. The development, release, and timing of any future features or functionality described for our products remains at our sole discretion. Performance is based on measurements and projections using standard IBM benchmarks in a controlled environment. The actual throughput or performance that any user will experience will vary depending upon many factors, including considerations such as the amount of multiprogramming in the user’s job stream, the I/O configuration, the storage configuration, and the workload processed. Therefore, no assurance can be given that an individual user will achieve results similar to those stated here.