What HANA on Power Means to IBM i

June 17, 2014 Alex Woodie

The recent news that IBM and SAP are working together to run HANA on Power piqued the interest of the SAP on IBM i community. While HANA won’t run directly on IBM i, just getting the in-memory data store on the Power Systems frame can bring new capabilities to organizations that run SAP on IBM i.

A week ago, we reported that IBM and SAP were getting serious about running HANA on Power and have started a test and evaluation program to analyze exactly what that venture entails. The program, which targets SAP’s preferred SUSE Linux distribution, is aimed squarely at the Power Linux environment, with the goal of giving customers another place to run HANA besides Intel X86-based servers and appliances.

IBM director of Power Systems product and offering management Steve Sibley says the program will involve “pre-general availability code” of HANA running on Power hardware. “Current SAP/IBM customers running their SAP applications on IBM Power7+ and Power8 platforms, and current SAP HANA customers, are potential candidates for the test and evaluation program,” Sibley tells IT Jungle via email.

We still don’t know what the timing of the program is, whether it’s already started, when it might end, or whether it will actually result in a finished product that Power customers can buy. There’s a lot of uncertainty at the moment, and the whole exercise could be squashed if it’s deemed the investment will exceed the returns. Neither IBM nor SAP are in the business of developing technology for technology’s’ sake.

But if what the folks at IBM are saying is true, and if Power on HANA can deliver a massive performance boost over X86, then there’s no reason to believe the players will not pursue the setup, particularly as their common competitor Oracle pursues its own unified in-memory stack.

At the moment, running HANA on Power looks good on paper. As IBM’s Sibley explains, Power runs more queries in parallel faster, across multiple cores, with more threads per core than Intel X86. “They also have increased memory bandwidth and faster I/O to ingest, move, and access data faster,” he says.

Sibley says HANA on Power will compete against X86 in both scale-out and scale-up environments. “In fact, based on preliminary results on POWER7+ [SMT4], the tested SAP HANA workloads should benefit greatly from the increased thread capacity,” he says. “We expect this to be even more dramatic with POWER8 [SMT8]. IBM Power Systems intends to be price competitive when the SAP HANA on IBM Power Systems becomes generally available.”

But what does this mean for IBM i shops? There’s a large and vibrant, if somewhat quiet, community of organizations that run SAP Business Suite on the combination of IBM i and DB2 for i. The work involved in moving a production workload from DB2 for i to the column-oriented HANA database probably wouldn’t appeal to these customers, even if HANA was supported on IBM i.

But using HANA for online analytic processing (OLAP) workloads on Power Systems is something that might appeal to the SAP on IBM i crowd. “IBM i clients often extend their environments with applications running on Linux or AIX and, while new enhancements to the IBM i Encoded Vector Indexing delivers enhanced performance for OLAP workloads, some IBM i clients may see a benefit of running SAP HANA for their OLAP workloads alongside their IBM i ERP solution,” Sibley says.

So while the full promise of HANA is that it’s a single data store that can house both transactional and analytic data and workloads, HANA as an OLAP engine for Power still carries a lot of possibilities. IBM offers its own column-oriented database for analytic workloads, called DB2 BLU, which will run on Linux and AIX on Power. So there’s some overlap of capabilities here. Using SAP’s BI tools to help manage and refine SAP-defined transaction data probably brings a set of advantages.

And who knows: Once HANA comes to Power for OLAP, who’s to say it can’t go the rest of the way and be used for transactions as well? A lot of organizations might jump at the chance to run a centralized, highly available and secure server that can house transactional and analytic data and workloads without the sort of administrative overhead and babysitting that accompany X86 server farms. This is definitely a project to keep an eye on.