Blade Servers Barely Nick IBM i Market

December 6, 2011 Dan Burger

After nearly four years of sales, IBM finds itself with a blade server for the IBM i-based Power Systems midmarket that is unwanted and as yet unable to build a following. Adjustments have been made, incentives have been provided, but this product still seems destined to be ignored by a large majority of users. Why? The answer is complex. This product is too complex compared to the servers the community is familiar with and has come to count on.

I don’t believe this is simply a case of not enough IBM i shops with a forward-thinking approach to IT. There are companies willing to innovate when they believe they’ve found the right product. Customer satisfaction in this community is legendary. Customer dissatisfaction is pretty strong, too. Running IBM i on blade servers in a BladeCenter has not found a comfort zone and pinning the reason for that on customers being comfortable in past instead of the future is inaccurate.

The dark shadow of complexity begins with the Virtual I/O Server, or VIOS as it is often shorted. It is not well liked among the i-centric propeller heads who have taken the time to examine it. VIOS is a product of AIX thinking, which is inherently more complex than is thought to be necessary on the IBM i side of the aisle. It’s a cross between logical partitioning and a bowl of spaghetti. To be comfortable with VIOS requires comfort with Unix. And although it is tempting to make this a religious war between IBM i and AIX factions, it’s not that simple. It’s really about what customers think is best versus what IBM thinks is the right path toward the future. And it’s about technology that is helping businesses rather than handcuffing businesses.

VIOS–as the driver for disks, controllers, and networking–gives the IBM i team a headache. Using VIOS made the IBM i BladeCenter easier for IBM by eliminating the need to create drivers for all the blade peripherals specifically for i5/OS and IBM i, but it isn’t the best or the simplest design. VIOS requires an uncomfortable mindshift, even for the innovative thinkers in the IBM i camp. It gets blamed on stubbornness, but it is rooted in building things right and not taking shortcuts.

Naturally, storage area networks (SANs) go hand-in-hand with blades, which are skinny on the storage front. IBM i customers are not just used to internal or directly attached disk, they believe the single-level storage technology on IBM i is superior. If they are going to use blades, they prefer the storage to remain on the i because it allows access to backups and expandable storage. As they see it, the IBM i is not just an element of the SAN. It is the sun in this solar system or is, in effect, its own SAN. Their reluctance to move to SAN storage is hardwired to keeping the lid on complexity.

In small to midsize shops, logical partitioning is not a total stranger. Yet, the idea of using it to create separate storage options draws disapproving looks even though it’s understood that Power7 delivers plenty of processing power, memory, and extra disk to host data for BladeCenter. It’s well understood that IBM i manages disk very well, especially in situations that most small to midsize customers experience–where it is spread across lots of disk arms.

This disapproving looks are because they know complexity when they smell it. In this case you have complexity related to VIOS, to SANs, and with potentially a much bigger networking job that drags in things like internal switches and cables, multiple power supplies, plug in modules. It not only smells like complexity, it smells expensive.

There’s no denying that the move to blades is an expensive undertaking when compared to traditional rack and tower servers. You have the chassis, the BladeCenter, all the connectivity, the physical site, and don’t forget–lots of power to run it. None of this is unique to IBM i. It’s the same with any operating system running on a BladeCenter platform and indeed with any other blade server. On a small scale, there are no savings with blade servers, even if there are operational advantages, because of the chassis and internal switching costs. Add more blades and the above costs get spread out and the total cost of ownership over several years somewhat soothes nerves that are easily rubbed raw as front-end costs escalate.

Companies where SANs are already in place along with the skills to manage them are in a better position to bring IBM i onto blades. They would also have much of the networking in place. But that’s not a common occurrence. Not yet anyway.

Bordering on the complexity that has already been mentioned is the requirement that Power-based blades can only run IBM i 6.1 and 7.1. Companies running V5R4 or earlier versions of the operating system are prevented from moving to blades, or any of the Power7 hardware for that matter. You can run V5R4 on the Power6 rack and tower servers, but not on the Power6 blades.

Unfortunately there is more than just a smattering of companies stuck on V5R4 because they have small but critical pieces of code that can’t make the program conversion required in the move to the newer versions of the operating system.

Larry Bolhuis, the well-known IBM i systems design consultant, recently told me of an idea he has for IBM to help those customers stuck at V5R4. I think it’s a good enough idea to share with you and maybe with enough voices it will get IBM’s attention.

Bolhuis is suggesting a miniature V5R4 emulation environment that would handle the small number of programs that in many cases are preventing companies from moving to new hardware where the majority of their programs would run on Power7 and either i 6.1 or 7.1. He’s already mentioned it to Steve Will, chief architect of IBM i, and has not been told it couldn’t be done. On the other hand, he’s not been told there is any interest in doing such a thing to provide an incentive for OS upgrades.

It could be a considerable investment for IBM to do this, but maybe not. It also could be something the ISVs would not be happy about. In a number of cases, the ISVs have customers over a barrel on this code conversion hang up because customers dropped software maintenance and are not on current releases. It’s one of those behind the scenes dramas that are playing out here and there.

One reason that blades look better now than they did early in 2011 is that the Hardware Management Console (HMC) was ditched. In its place is the Systems Director Management Console (SDMC).

Although VIOS is not well liked (one of Bolhuis’ customers refers to VIOS as “the world’s worst command line), if you’re going to do blades you’ll have to get used to it. The HMC went a long way to make that proposition very unlikely. The late Al Barsa used to refer to the HMC as the “Hardware Mangler Console.” Its best features, reliability and flexibility, were battered and bruised because it had no GUI and no Web interface. It was harder to use than a short-handled shovel.

Of equal if not greater importance, Bolhuis told me, was HMC’s inability to provide Virtual I/O Server redundancy. “In instances where VIOS is supporting multiple partitions, the HMC created a single point of failure,” Bolhuis says. “That’s not good. You can’t take it down for maintenance. Dual VIOS supported by SDMC makes it so the work is shared and the IBM i sees two ways to get to the data. It’s the same data twice, but IBM i and VIOS understand that. And if one VIOS server goes away, the other can do all the work for a while.”

Bolhuis is somewhat kind when talking about VIOS. He describes the day-to-day operation of VIOS as “not terrible if you are not changing disk.” He warns, however, that service packs will occasionally need to be added and that most companies will probably want to hire an expert for that task. “It’s not like service packs on IBM i. You don’t just slap in a CD and go,” he notes.

I also talked with worldwide PHP evangelist Mike Pavlak recently on the topic of blades in IBM i shops. Pavlak estimates that blade servers are operational in one out of every ten shops that his employer, Zend Technologies, contacts. In his opinion, companies are contracting with business partners to do all the set up that causes fits for unskilled laborers. By his reckoning, the number of IBM i shops that operate without a systems administrator has risen dramatically during the past 10 to 12 years and he expects that number to increase as companies draw closer to trusted business partners for the skills and expertise associated with new and complex technologies.

Meanwhile, Bolhuis, who is one of those trusted business partners, sees little traction for blade servers in the small to midsize companies that his company, Frankeni Technology Consulting, works with.

“I was involved in an early beta, actually pre-beta, program at IBM to make blades work on IBM i. All of us on the team were i certified and those of us who worked directly with IBM i shops asked ourselves whether this would fly in our customers’ shops. It had the capability and it’s cool, but most companies want that simple box that they can walk in and push the button after a power outage and it fires right up.”

A quote attributed to a Larry Bolhuis customer claiming the HMC was “the world’s worst command line” was incorrect in this article as originally published. The customer used that description to describe VIOS instead. IT Jungle regrets the error.