What To Do With All Those Spare CPWs

If there’s one thing that the average IBM i shop with a typical business forecast doesn’t need, it’s more CPW. The latest generation of Power8 servers offer more than enough computational oomph to get the job done, which raises the question: What should one do with those spare CPWs? With some forward thinking, the average IBM i shop maybe doesn’t have to be so average anymore.

There is no doubt there’s a power glut of Power8 iron in IBM‘s IBM i market. Each Power8 chip has roughly 50 percent more computational power per core than the Power7 chip launched in 2010. And since IBM is adding more cores (12 in Power8 versus eight in Power7) to its systems, the horsepower quickly climbs.

So when you consider that IBM is making similar investments in keeping ahead of Intel in the memory bandwidth department, you find that an average, run-of-the-mill P20-tier system like the Power E850 has twice the system throughput as a similarly sized P20-tier Power7 box, the Power 750.

The problem is, the workloads of IBM i shops aren’t growing this fast–not by a longshot. Thanks to the relative efficiency of RPG and COBOL code, ERP systems, and other business applications that were developed decades ago to run on AS/400 and iSeries systems that were computationally constrained (relatively speaking) run with plenty of headroom on new systems. (This wouldn’t be such an issue had we all moved to Java, of course). It’s no wonder so many shops are downgrading their hardware footprint when upgrading their systems. The smallest Power8 systems are plenty powerful for most IBM i shops.

While IBM fights to retain the premium that IBM i shops have historically paid over AIX and Linux customers–and attempts to prevent shops from downgrading from enterprise-class systems into cheaper midrange-class systems–it begs the question: What will IBM i shops do with all those CPWs if their database and ERP workloads don’t need them?

Here’s one wild and crazy idea: do some analytics.

That may sound totally nuts to the average IBM i shop, which may do some reporting on the system but nothing that would be considered “business intelligence.” But the truth is, the latest incarnation of BI–big data analytics–is coming to a theater near you, and it’s coming sooner than you think.

The biggest companies in every industry are using big data tech to gain a competitive edge. These adopters often built their own solutions, but today the market for big data tech is flattening, the solutions are becoming more shrink-wrapped, and commodity forces are at work. That’s great news for the smaller fish who live downstream from the big guys.

The question that IBM i shops should be asking themselves isn’t whether they need to have a big data strategy (they should). Instead, the right question is: How should they get started?

The big data market today revolves around the commodity Lintel system. Much of the big data tech today–the distributed file systems and schema-less data stores–was developed by Silicon Valley Web giants like Google, Yahoo, Facebook, and LinkedIn, and then made open source. That’s good news for IBM, which has worked hard to retrofit some of the latest and greatest big data tech, such as Apache Hadoop and Apache Spark, to run on its Power processors, and to be supported on Power Systems and OpenPower ecosystems.

While IBM is playing catchup when it comes to big data tech on Power, it does have a few aces up its sleeve, including superior memory bandwidth compared to Intel’s architecture. When big data gets really big and you want to access it quickly, that speed turns into a concerted advantage.

Most, if not all, of the big data tech that IBM is supporting runs on Linux. This brings us to some other good news: thanks to IBM’s work on solving the Little Endian/Big Endian issue, much of the software that’s been developed to run on Linux for X64 can also run on Linux for Power without too much work. On its website, IBM says that Canonical, the company behind Ubuntu Linux, managed to port 40,000 Linux apps to Power in just 160 days.

In addition to requiring lots of storage, big data analytics also eat CPUs (and their EBCDIC cousin, CPW) for breakfast. That’s good news for Power Systems users looking for useful things to do with their big, expensive iron while it’s sitting idle. All it takes is some courage to start experimenting–and of course a couple of Linux LPARs to start playing around.

Neo4j uses IBM’s Power8 technology to take big graph analytics to another level.

One of the most interesting big data-related developments to happen in the IBM Power world lately is the partnership created between Neo Technology and IBM last October.

Neo, if you’re not familiar, is the leading developer of graph databases for big data analytics, and its flagship product, called Neo4j, is being used in hundreds of blue-chip companies such as Wal-Mart, Pitney Bowes, and UBS. Just as Facebook uses a graph to connect systems, Neo4j can quickly find similarities among billions of connected entities, and the software runs at the heart of a variety of applications in fraud detection, master data management, and real-time recommendation generation.

The cool thing about Neo and IBM teaming up is the capability to solve truly massive graph problems on Power8 servers. The companies did the work to enable Neo4j to run directly atop the Coherent Accelerator Processor Interface (CAPI) cards that IBM uses to connect co-processors, such as Field Programmable Gate Arrays (FPGAs) and Graphic Processing Units (GPUs) to the PCI bus. Neo executives say they’re able to access up to 40TB of CAPI-supplied flash memory as if it were main memory, bypassing Linux and the file system in the process (see graphic above).

This enables Neo4j to build and query graph databases that are 10x bigger than anything possible on X64, according to Neo–graphs measured in the hundreds of billions to trillions of nodes. Fraud detection, bioinformatics, and analytics for the Internet of Things (IoT) are the top workloads expected to be enabled by this technology, which will become available later this year.

OK, so you’re not going to casually do some entity analytics against a trillion-edge graph using the spare capacity in your Power8 system. You’ll probably buy a dedicated system to do this work. But the point is that these sorts of big data analytic capabilities are more accessible than you may think.

After all, IBM is adding support for semi-structured formats, like JSON, in DB2 for i and other DB2 flavors. What will it do next? Chances are good, some sort of data analytics is near the top of the list, so it may be a good time to start familiarizing yourself with the technology that’s available, and how you might better leverage your existing data–and your spare CPWs–to compete better in our uber-connected, big data world.

RELATED STORIES

Power8 And The Potential Oomph In Midrange And Big Boxes

Power Chips To Get A GPU Boost Through Nvidia Partnership