Delivering Persistence Without Compromising Performance
Asim Mustafa Khan discusses IBM's commitment to Persistent Memory Enhancements.
By Asim Mustafa Khan08/01/2019
This blog is part of the strategy that discusses the memory enhancements on IBM Power Systems for SAP HANA.
Later this year, IBM intends to enhance our advanced virtualization platform (PowerVM) with a new feature called PowerVM with Persistent Memory(PMEM) that will address the following business requirements that clients have shared with us:
Clients demand choice, and all x86 vendors who are embracing a single persistence technology from Intel are, in essence, the same proprietary technology.
Our current SAP HANA on IBM Power clients do not have nearly as many server outages, but even they would like to be able to restart faster after patching their software stack.
Clients (on x86 or Power) do not want to tradeoff performance for persistence.
Clients want to add the persistent memory capability without having to rip and replace or purchase expensive add-ons to hardware they just recently bought.
PowerVM with PMEM (Persistent MEMory) is an enhancement in our virtualization platform that will create persistent memory volumes using existing DRAM, maintaining data persistence across application, operating systems and partition restarts. This capability requires no changes to existing applications.
Persistence Without Compromising Performance, Virtualization and TCO
Since this feature is based on existing DRAM technologies, it has the same performance characteristics clients already experience today, providing clients with the peace of mind of using technology that already meets their performance requirements.
You may be wondering why we think that persistence at the LPAR level is the best solution. We could have aimed to match what the x86 competition was planning, but the loud feedback we received from our existing HANA on Power clients was: “Why try to solve a problem you don’t have?”
The table below indicates some of the typical comparisons that clients may want to use when comparing persistency related solutions. In particular, I’d like to highlight that IBM PowerVM with PMEM will shrink downtime window significantly—even more than other technologies by drastically reducing shutdown time, as well as startup due to technical capabilities only available on IBM Power Systems.
Clients who had lots of experience with x86 and POWER for SAP HANA told us that, when they switched to POWER, hardware related outages became very rare (unlike when HANA is running on x86 based servers). When using POWER, the vast majority (if not all) of the outages that the clients have are for operational needs like OS or application related upgrades and patching. And they needed to reduce this maintenance window to improve availability of the environment to their businesses without the fear of degrading everyday performance. Our PowerVM with PMEM solution meets and in fact exceeds all these requirements.
Unlike x86 vendors, we won’t force our existing SAP HANA on POWER9 clients to rip and replace their recently procured hardware and pay to have this feature. It will be available to all clients running POWER9 processor-based servers with a simple software update at no additional cost!
Since this feature will be supported with virtualization on Power Systems, clients will continue to have the same flexible scalability and granularity they already enjoy along with dynamic and flexible CPU and Memory resource assignments to each VM on the system. If the clients want to deploy an x86 persistent solution, it will have to be bare metal implementation for all the modules, as virtualization using Optane is not yet supported by SAP and that will make the deployment on x86 hardware less flexible and significantly more expensive.
We believe clients running SAP HANA on POWER will continue to have significantly better TCO than our competition without the trade-offs that are forced by their proprietary persistent memory technologies.
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.