Virtualization on i Boxes Depends on Consolidation, New Workloads
May 18, 2009 Dan Burger
Server consolidation has been on the minds of IBM executives for years. It’s been talked about like it’s some kind of big deal since almost 10 years BC. (That’s Before Convergence.) Usually it’s been described as a top priority for companies with multiple AS/400 boxes spread across wide geographic areas. During these virtualized times, when the X64 server farms are catching on to this technology, it’s appropriate to take a closer look at what virtualization brings to the i platform.
First of all, it really has been a pretty big deal–at least in the higher end shops where the bigger Power Systems i boxes work each day and night. At the COMMON 2009 Annual Conference and Expo in Reno, Nevada, last month, Ian Jarman, manager of Power Systems software, explained his view on the virtualization/server consolidation market, and last week IT Jungle checked in with Ian Robinson, PowerVM offering manager, who added his perspectives. Here’s what the two Ians had to say.
“When I go to customer conferences and ask how many people are using virtualization in the i community, it is the majority,” Jarman says. “There may be a lot of older AS/400 systems out there (not using virtualization), but on the modern systems–Power5 and Power6–I think the majority of i clients are using virtualization. If you look at the midrange and larger systems–the 550, 570, and 595–it is definitely the majority. If I was to talk to the Large User Group, it would not be the case of whether they have partitions, it would be whether they have tens or hundreds. The conversation has moved from should you deploy it to how do you manage it.”
The reason IBM introduced logical partitioning on the AS/400 in 1999 with OS/400 V4R4 was because many of the enterprise customers were ready to consolidate their AS/400 footprints into a virtualized environment. These early adopters of virtualization had distributed deployments with AS/400s in data centers for each country’s operations or in branch offices in many locations. Much of the blame for these distributed AS/400 environments can be attributed to a communications infrastructure that, at that time, was not conducive to running servers centrally. And it was expensive, too.
As the communications infrastructure improved and telecom companies starting charging reasonable prices for bandwidth, consolidation became much more appealing, Jarman says. Companies found real savings when buying and managing fewer servers. Rapid deployment of new applications and operating system upgrades and avoiding the customary server installation hassles were server management features that benefited all the distributed locations, much like the X64 world is recognizing today. The math is pretty simple. When you have hundreds, or even thousands, of systems that need updating, a quick comparison of performing that task versus updating three partitions that are serving 1,000 locations is a no-brainer.
Robinson points to workload consolidation when he describes the value in virtualized environments.
“For customers buying large systems,” he says, “the best bang for the buck out of a hardware investment is using multiple i OS partitions. They are using it to consolidate a large number of workloads from many machines, including, in some cases, Linux X86 and X64 boxes. Virtualization is an enabling technology that allows a lot of solutions. The primary one that most use is consolidation. This makes sense when scaling up to a big machine with a lot of cores and memory. Then you can put tens or hundreds of workloads on a machine.”
Will users of the smaller Power Systems i boxes take to virtualization? Robinson says that in i shops virtualization is not used as much for consolidating workloads on smaller systems, but it is being deployed for things like improving recoverability and because it is easier to scale.
“If you want to provide backup for other hardware, it can be done more easily in an LPAR,” he says. “Some customers are looking at that using the i box to create a virtualized version of Linux X86 workloads in case those machines go down. It provides a live, up-and-running copy, even if it not going to run permanently on the i. It boosts the availability of those Linux workloads.”
This is not a primary reason for people to use Linux, Robinson admits, but it addresses a need for companies using virtualization to become more resilient on the entry-level machines.
However, the topic of Linux workloads does not pass without the mention of 5,700 ISV applications that are natively compiled for Linux on Power. That number is growing every month. Robinson, who is quite the Linux evangelist, says that potential for moving compiled X86 workloads onto Power Systems iron where they can run natively atop the Power LxVM emulation environment that IBM created using the QuickTransit software from Transitive. (IBM bought Transitive last November, and it is that company that brought Robinson to IBM.)
Another scenario Robinson uses to illustrate the advantages that Power Systems and their logical partitions hold is when an application needs to scale. He likes the example of an e-commerce application that has usage rates that spike seasonally. In this case, the workload could be temporarily replicated to deal with the spike and it could be shutdown after the activity subsided.
Despite the encouraging words from Jarman and Robinson, I have to believe that if IBM was having great success with virtualization on the Power Systems i it would be promoted more enthusiastically. But, then again, what part of the i story does Big Blue promote enthusiastically? Maybe they raise the roof at some of those Large User Group events. Since the i and the p systems have converged, there is decidedly less platform-specific talk and more emphasis placed on Power Systems. This fuzziness doesn’t make it any easier to identify where trends are occurring.
Neither of the Ians will talk about statistics that quantify the use of PowerVM running on the i side of the house. They both season their comments with terms like “strong demand” and “increasing numbers” and “a lot more activity.” One stat that has been used recently to support a demand for virtualization is that 65 percent of Power6 systems have PowerVM shipped with them. No breakdown of how this applies to boxes shipped with i OS versus AIX is provided, but Jarman says the AIX side has more to do with this than the i side–where partitioning is not new.
Even though IBM is going to track core activations because they relate to licensing, the two Ians stand tight-lipped when it comes to identifying platform-specific trends. “It’s hard for us to measure that because we don’t call them System i or System p anymore,” Robinson says.
OK. We can’t really track the past with anything that resembles a sharp instrument, so let’s just move to the future. Not surprisingly, there is a PowerVM roadmap.
“The most exciting part of the roadmap to me is in the area of cross-platform virtualization,” Robinson says. “IBM did a great job of bringing together the i and p communities. We now have a common hardware platform, yet we still have the OSes in their own partitions with their own libraries of applications available. This is all part of the evolution of the Power Systems architecture to be an enterprise consolidation platform.”
Approximately a year ago, PowerVM rolled out an X86 Linux runtime environment for Power Systems. Robinson–the above-mentioned Power LxVM–hinted that similar products will be introduced so that Power Systems “will ultimately run just about any enterprise workload.” In addition to Robinson, other Transitive personnel were absorbed into IBM, so the engineering skills in cross-platform virtualization are in the mix not only on the development side but also on the sales side where they help educate customers, resellers, and field sales.
The Power Systems roadmap, and particularly the PowerVM roadmap, quietly emphasizes additional workloads that will benefit customers of the i, as it puts the Power Systems out front. According to Robinson, we (the AS/400 community) should count on scaling up to bigger systems and rolling in new workloads, particularly when companies make a strategic decision to narrow to one architecture using one box, one VM format, and managed through one console.
“There’s not too much meaning to platforms anymore,” he says. “We have hardware that ranges from the 520 or the blades and up to the 595. All of which can run i in partitions. In one sense, the range of machines that i customers have to choose from is bigger and better than ever.”