A Bit More Blegging

by Administrator on June 5, 2008

I love it when a new word is coined in the blogosphere.  This one comes from Tony Pearson over at IBM:  “blegging.”  A “bleg” is, I guess, a blog post that is asking for something from the reader, as I did with my recent de-duplication questionnaire for vendors.  Tony, I am going to bleg for something again.  (Thanks, by the way, for your response to the de-dupe questionnaire.)

This time the bleg is for your assistance in clarifying a few issues about virtualization using LPARs on an IBM z10.

I am, as you know, favorably predisposed to use LPAR technology, rather than x86 with extents and hypervisor software, to consolidate and virtualize servers and I have written about the value proposition both here and in mainstream pubs.  The problem is that some of what I learned from chats and formal interviews and press releases out of Big Blue, and reported in the trades, has caused a ruckus from readers.

Press materials around z10, for example, stated that one could use LPARs and virtualize up to 1500 x86 servers at a fraction of the cost that it would take to do so with VMware.  The exact number quoted was “$660 per VM.”  Compared to doing the same with VMware, that sounded like a great bargain.  A reader now challenges this number, stating that z/OS only supports 60 LPARs.  So, where is the 1500 number coming from?  Good question.  So, I am blegging for an answer.

Second, IBM jocks told me that Sysplex provides a much more robust guarantee against a VM meltdown by enabling load to shift to other processor modules if one fails.  This is, of course, much more resilient sounding than a VMware hosting metaphor.  Except, the reference to Sysplex, a reader suggests, must have been metaphorical, since z/OS and z10 don’t feature Sysplex technology.  I bleg for some clarification on this point.  (The fellow also said that while z10 supports multiple processor modules, he didn’t think that the cost for this technology was part of the dollars analysis that went into the “1500 servers for $660 per” statements made by IBM.  Does this sound accurate?)

Finally, I had noted that with tools such as DFSMS and DFHSM resource utilization efficiency was better behind a mainframe than behind open servers.  The reader gently reminded me that DFSMS and DFHSM had not been fully ported to zOS.  Where does this stand?  Are their other facilities in the works for zOS that are comparable to SMS and HSM?  I bleg to know.

I have directed these questions to IBM through analyst relations channels, but have received no response.  Since you seem to be the most responsive IBMer I know, can you clarify this for me, Tony?  As I said, IBM’s VMware alternative sounded great to me, and I want to believe.  But, this confusion is making that difficult.

{ 9 comments… read them below or add one }

RC June 5, 2008 at 8:13 pm

After a few minutes of poking around the IBM site, I think I see how its done.

At first, I thought the z6 Grue had eaten an x86 ISA. IBM isn’t touting any such feature, so it must be something else.

It looks like they are basing the virtualization numbers on moving Linux systems from standalone hardware or VMware to running on z/VM.

Quoting page 2 of [this PDF downloadable via FTP from IBM]

Flexibility to create as many as 60 LPARs on the z10 EC and z9 EC
Ability to virtualize each LPAR into hundreds or more virtual machines

Sixty times hundreds is sixty-hundreds.

Even though I don’t find “binary” or “ABI” anywhere in the pdf, there is a mention of RHAS4 and SLES9. Both of these distros are available for zSeries, and would guest in z/VM.

Without the x86 ISA in microcode, I wouldn’t expect to run thousands of copies of Vista on zSeries anytime soon. (Shudder.) But then again, maybe that isn’t such a bad thing?

Administrator June 6, 2008 at 8:26 am

Thanks, Robert. It is still beginning to smell of fun with numbers to me. I am looking forward to IBM’s “official” statement.

popadin June 6, 2008 at 9:36 am

Seems to be some confusion about how to get 1500 zLinux instances out of 60 LAPRS and of course you can’t do that. An 60 LAPRS will support z/OS or z/VM and it will support specialty engines such as IFL, ZIFSbut only one per LAPR. a z/VM instance will support all kinds of zLinux instances but zLinux runs in an IFL which is something you have within an LPAR.

Not sure about the DFSMS and HSM “not fully ported to z/OS” statement, DFxMS is part of the OS, no porting required.

RC June 6, 2008 at 12:16 pm

1500 instances at $660 each equals approximately (Dr Evil Voice) One Meelion Dollars.

Weird total that. I’m pretty sure if you wave $1M at IBM, they will make a machine appear that will run 1500 instances of Leenux. It may be a Dead-Parrot distro, but once you recompile your Web 2.0 App, it should run.

1) Build time machine.
2) Travel back to 1988.
3) Tell Bill Joy to keep the C compiler bundled with SunOS, keep making the Sun486i, and stay away from SYS V.
4) Hire a henchman to steal the Solaris manual that Linus based his kernel on.
5) Profit!

zax218 June 6, 2008 at 12:34 pm

I guess my question is, How many companies have large Linux deployments on VMWARE? If this is supposed to compete with VMWARE, I would think that it would target VMWARE’s primary market. Windows Servers and Desktops.

John, you continue to tout this box as a VMWARE killer, but the reality is that it doesn’t even address VMWARE’s primary market segment. Maybe it would be more appropriate to position this against Sun’s virtualization effort or RH/KVM on Commodity X86 HW.

Administrator June 6, 2008 at 3:18 pm

I am not calling IBM a VMware killer. I am asking, as everyone should be, whether they feel that x86 with extents provides a sufficiently resilient or the most cost effective platform for vrtualizing server environments. Based on their public claims, 1500 servers virtualized for $660 per server, in a Z10, I was inclined to believe that a mainframe might provide a more resilient and cost-effective solution for companies that were virtualizing a lot of servers. Now, I am not so sure.

I am blegging IBM to respond to what intelligent consumers (and zOS users) have pointed out to me might be flaws in the IBM marketing around this idea. I wouldn’t be asking if I was sold on this method.

I have already said that, given IBM’s lack of support for Microsoft OS virtualization, and given Microsoft’s preeminence in contemporary server deployments, that IBM’s virtualization story may well be flawed. I just want to know if they sold us a bill of goods when they made their claims about the efficiencies and the capabilities/capacities they represented in their claims.

Shouldn’t those issues be explored? How does that translate into a view that mainframes are VMware killers? I don’t know what you are seeing.

Frankly, right now I am turning my attention to Hyper-V and Virtual Iron. Hyper-V seems to be very Microsoft application friendly, but it obviously lacks the additional capabilities that VMware or VI bring to the party. VI is much less pricey than VMware, and I am setting up a test environment in my labs currently to do a feature to feature comparison between VI and VMware ESX as well as a cost-comparison.

I guess anyone who drinks the VMware Kool Aid won’t want to know the results. I guess doing the test will make me out to be saying that VI is a VMware killer. Nonsense!

In retrospect, I was very keen to see IBM provide a rock solid virtualization story with its mainframe. It bothers me, however, if the supporting claims for their arguments prove to be deceptive. I will readily recant my initial support for their platform if it is discovered that they prevaricated in their marketing claims. That criteria is applied equally to all vendors, including VMware, which routinely deceives consumers about its capabilities in the presentations I have witnessed.

az990tony June 9, 2008 at 8:07 pm

Hi Jon,
Thanks for the kind words. I’ll investigate and get back to you.
– Tony

az990tony June 12, 2008 at 1:00 pm

Hi Jon,
I was able to speak to the guy in IBM who did the calculations, and provide all the details here on my blog:


– Tony

Administrator June 15, 2008 at 6:06 pm

The Sun fellow who first posed the potential discrepancies in the IBM numbers I cited in my Mainframe Executive column and here has posted some data of his own. Worth a read.

Previous post:

Next post: