IBM dives into converged hardware with PureSystems

Whatever you do, don’t compare IBM’s new PureSystems to Oracle’s Exadata or Cisco’s Unified Compute System.

IBM spent three years and $ 2 billion to develop these systems, said Rod Adkins, SVP of the company’s Server and Technologies Group. That’s no knee-jerk reaction, but it is a bit late to the party.

In Adkins’ view, PureSystems split the difference  between single-purpose appliances a la Netezza, Exadata et al. and general-purpose hardware. Treading that fine line between tight integration and flexibility is a neat trick and one that Cisco claims to have accomplished with UCS.

Mix-and-match compute nodes

In IBM’s case, however, the buyer can actually mix compute nodes — X86 and/or Power7 based — in the same enclosure and even run different operating systems — AIX, Linux, Windows — as needed. ”No other system offers consolidation without migration — we designed it so it can deal with multiple compute nodes in the same box,” Adkins said in an interview.

By packaging all those workloads together and by nixing the top-of-rack switch used in some converged hardware, IBM was able to slice latency times — in some cases it claims by 50 percent — to boost application performance. Eliminating that switch cuts out a lot of the data round trips and boosts performance, Adkins said.

There are two product lines. PureFlex is the infrastructure foundation piece (the compute node plus 10GB network switch; IBM V7000 storage etc.) PureApplications will layer IBM DB2 database, WebSphere middleware and other software atop that infrastructure. PureFlex is due this quarter, PureApplications will come out later, he said. Both will come in three SKUs — Express being the smallest, Standard and Enterprise.

The company is positioning the boxes — which will run IBM’ SmartCloud — as a good way for companies to bootstrap their compute clouds. “With the SmartCloud integration you can basically do a four-click deployment to provision a new cloud — you identify the resources, pool them, pick the applications, then deploy and run,” Adkins said.

Richard Ptak, co-founder of research firm Ptak Noel & Associates, especially liked the fact that IBM has worked with 100 ISVs to package up their applications so they will be easy to deploy with the required resources for cloud use.

Reviving scale-up vs. scale-out debate

Lateness is one thing. There’s also debate whether high-end and pricey boxes like these (PureFlex starts at $ 100,000 for the low-end Express model) are the right solution for cloud computing. Conventional wisdom holds that the best way to attack cloud computing is with massively scaled-out commodity hardware.

But Ptak said PureSystems’ flexibility could make a difference.  ”You can mix and match multiple blades and move the virtual environment from one blade to another if needed — and all of that can be managed,” he said.

The net result, according to Ptak, is that PureSystems give users the simplicity of an appliance, the flexibility of the traditional computing model and the automatic scale-up/scale-down of the cloud computing model.

That sounds like computing nirvana if it works as advertised.  Still, it’s hard not to see this as IBM’s response to what Oracle, Cisco, even what the EMC-Cisco-VMware vBlock consortium did. What IBM has done may be slick but many cloud customers may not see a lot of use for high-end converged hardware. Especially when they can yoke together lots of cheap servers to run their cloud or — more likely — turn to Amazon or Rackspace to do it for them.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

  • Forecasting the future cloud computing market
  • Big data: beyond analytics
  • New challenges for the IT organization



GigaOM