POWER6 Goes Thud: Part VI

IBM's Me-Too-But-Not-Now Virtualization Story

Here's an old joke I heard a long time ago. Three children are playing in the yard. The first one says, "My mommy's a lawyer and she gives me ice cream when I'm good and I love that." The second one says, "My mommy's a doctor and she gives me healthy fruit when I'm good, and I love that." The third one says, "My mommy works for IBM and she just tells me how great it will be the day she decides to feed me."

I was reminded of that joke recently when I saw the lengths to which IBM will go to spin their catch-up virtualization story as innovative and ready for you today. This is particularly odd when you consider that IBM introduced virtualization to the world 35 years ago with VM. But hubris affects all great companies, and IBM is no exception. To hear them tell the story, everything from fully isolated workload partitioning to on-the-fly live migration from one virtual pSeries system to another is here, just for the asking, now that the fabulous POWER6 has arrived.

For the record, here are IBM's current offerings for the POWER5- and POWER6-based systems today, listed according to Gartner's virtualization categories:

"But wait!" I hear you cry. "Haven't you heard of IBM's new workload partitions (WPARs) in AIX 6.1?" Indeed I have. In fact, IBM's Magical Mystical Marketing Machine is trying to get customers to think of little else. WPARs will slide into that third column under "OS Virtualization" as soon as Mother IBM decides to serve it up, probably when AIX 6.1 ships (possibly) by the end of this calendar year. WPARs are IBM's attempt to catch up with Solaris containers, introduced in Solaris 10 two and half years ago. This itself is odd since IBM exhausted a lot of breath over most of those two and half years telling customers what a bad idea Solaris containers were. Now they're flooding their customer base with beta WPAR code and telling everyone how absolutely essential OS virtualization is. [I must tip my hat to my IBM brethren here, though. They did come out with a glowing review of Solaris containers (and much else about Solaris) in a recent Redbook, now that IBM has become a Solaris reseller.]

Now IBM is showing off its first stab at a Live Application Mobility feature with WPARs (see the details in IBM's Redbook). Customers who have seen the demo in Austin report that it looks impressive under laboratory conditions, but as with all pre-release code, they're certainly not willing to put it in production yet. Nor do most have any operational policies for live migration of large mission-critical applications, as they do for small ones with VMotion and maybe Xen, and aren't likely to any time soon. This makes a lot of sense: WPARs have none of the real-world production bake-in of Solaris containers and adding on-the-fly migration on top of that immaturity is a huge amount of new code to add to a first release of a technology. Customers may also have a problem with the fact that there's no ISV support for WPARs at the moment like there is for Solaris containers. Don't expect that to stop the Blue Spinmeisters from telling you that WPARs are a leading new technology, were all IBM's idea, and represent massive new code that works flawlessly.

Of course, IBM has not publicly said how much all that virtual goodness is going to cost you when it eventually arrives. If it's anything like the way they gouge customers with LPARs, many customers may judge this whole WPAR thing unworthy of the price.

Suffice to say, Sun is developing checkpoint/restart and live migration capabilities for its current and future SPARC products, and I daresay that you may just see the first of those right around the time AIX 6.1 is finally production-ready. They will be a nice addition to our already industry-leading SPARC virtualization options, shown below:

  • Dynamic System Domains (DSDs) on Sun Fire and Extended System Boards (XSBs) on Sun SPARC Enterprise: These are electrically- and fault-isolated partitions on Sun's midrange and high end servers. Hardware faults in one domain cannot propagate to other domains due to the severing of electrical connections on the centerplane between system boards. CPU, memory, and I/O resources can be migrated from one domain to another without rebooting either domain.
  • Logical Domains (LDOMs) on Sun Fire T1000/T2000, Sun SPARC Enterprise T1000/T2000: These are hypervisor-based entities based on the sun4v architecture, which exports subsets of physical resources to separate instances of the OS. Up to 32 LDOMs are supported on the T1000 and T2000 platforms. Currently, Solaris, Ubuntu Linux, and FreeBSD all run in LDOMs.
  • Solaris Containers: Based on the theory that the best virtualization engine in computer science history has been the operating system itself, Solaris containers are extremely lightweight resource boundaries within a single Solaris image. Solaris containers build on the 15-year legacy of isolation technology from Trusted Solaris. Each container has its own root filesystem and IP addresses, and is such a good isolation mechanism that even Oracle recognizes a Solaris container as good as a hard partition for licensing purposes. Yet Sun has tested hundreds of containers in a single OS instance with no more than 5% additional overhead in the system, and usually a lot less.
  • Solaris Resource Manager: This is Sun's fair share resource scheduler. System administrators can manage CPU, memory, and I/O allocation by user, group, or project, all within a Solaris container.

Although this is a SPARC-oriented blog, I would be remiss if I didn't mention the fact that the OS-level features, Solaris containers and Solaris Resource Manager, are also available to the two-thirds of Solaris 10 licensees (over 8 million and counting) who run it on hundreds of commodity x86/x64 platforms, including IBM's own. AIX does not run there at all.

To be fair, Sun and IBM implement features differently in hardware and software, and virtualization is not an end in itself, merely a means to the end of flexible resource management. So let's take a look at the resource management features themselves and how and when Sun and IBM first implemented them:

(WHAT, not HOW)
First Introduced
First Introduced
IBM's Lag Time
On-line Dynamic Resource Re-Allocation
Within an OS Instance
Solaris Resource Manager,
various Solaris mechanisms,
Workload Manager,
various AIX mechanisms,
Let's call it a wash
On-line Dynamic Resource Re-Allocation
Between OS Instances
Sun E10000 DSDs
Solaris 2.5.1
AIX 5.2
6 years
Virtual OS Image in
Same OS Instance
Solaris Containers
Solaris 10
January 2005
AIX 6.1
At least 3 years
Checkpoint/Restart or Migration of
Running OS Instance
Solaris 10
AIX 6.1
Electrical Fault Isolation
Between OS Instances
Cray CS6400
Solaris 2.3
NEVER Forever?
On-line CPU/Mem/IO Replacement Sun E10000 DSDs
Solaris 2.5.1
NEVER Forever?

Given this years-long history of following Sun's lead in UNIX system resource management, IBM's Magical Mystical Marketing Machine has been on a rampage to spin their situation this way: Sun is lagging IBM in virtualization technologies. (Huh?!?) I guess they figure that the best defense is an offensive Marketing Department. Don't let them snooker you. If an IBM rep tries out this ridiculous story on you, put this chart in front of him and ask him: (1) why is IBM always so far behind Sun; and (2) why do they always cost so gosh-darned much?

All of Sun's virtualization and resource management technologies can be mixed and matched: you can use Resource Manager within a Solaris container within a Solaris instance within a DSD, XSB, or LDOM. And of course, unlike IBM, all of Sun's virtualization technologies are available and fully functional with the purchase of a server at absolutely no additional cost to you.

This last point is especially important to remember when you realize that IBM's ultimate strategy is to virtualize everything in your datacenter, not with technology, but with these guys. IBM Global Services is the oldest, most expensive, lowest-tech solution to your resource management problems. Do you know how hard it is to get rid of them once they're in your datacenter?

Virtually impossible.


IBM®, POWER™, POWER4™, POWER5™, POWER6™, pSeries®, System p5™, Redbook®, DB2®, and AIX® are all trademarks of the International Business Machines Corporation in the United States, other countries, or both.


As far as I remember, IBM didn't invent virtualization. VM was created at an University to avoid paying IBM software licenses, then IBM bougth the whole development... LPARs were created by Amdall, and then copied by IBM... Please correct me if I'm wrong.

Posted by Matias on August 28, 2007 at 07:50 AM EDT #

Post a Comment:
Comments are closed for this entry.



« December 2016