Wink and a Nod

by Administrator on March 14, 2008

Sorry to be out of touch of late.  A heavy workload and lots of travel have kept us quiet, but also begun filling the coffers with more data and user stories to tell. 

This post is about mainframe-centric virtualization, something that we have been getting a lot of inquiries about since the announcement at SHARE a couple of weeks ago about IBM’s new z10 mainframe.  Here’s the skinny:

z10 is the high end of the growing line of superservers from Big Blue.  For about a million bucks, a company can deploy one of these and use it, among other things, to consolidate and virtualize upwards of 1500 x86 servers using logical partitioning technology that has been refined over the past 25 to 30 years.  This isn’t really news — the virtualized environment at least — since I can recall consolidating UNIX servers onto mainframe partitions back in the 1980s.  What is very interesting about it are two things:

First, considering that the preponderance of server virtualization via VMware, etc. that is happening today involves mainly file servers and web servers (despite what VMware evangelists say, comparatively few database servers are being virtualized owing to I/O constraints inherent in VMware over x86 architecture), large companies might be seeing the sense in using mainframes for the job.  The costs work out to about $600 per VM, versus 10 to 20 VMs riding on a $20K x86 server box with $3K per VM for VMware licenses when you do things the way VMware wants you to.  That does NOT include the costs for trained personnel to create, implement and maintain the respective environments, which might favor x86 approaches from a short term perspective (CAPEX plus OPEX), but quickly level out or favor the mainframe solution when you are talking about 200+ servers. 

A compelling argument could even be made that mainframes are greener than the x86 racks they replace.  The math is simple.  Numbers of x86 environments that can be consolidated into one high end x86 server (and I am being generous here): about 20.  Numbers of x86 environments you can consolidate on one z10:  1500.  The z10 occupies the equivalent space and consumes the equivalent power of two server racks.

The other key point worth making about this scenario is that storage behind a z10 must conform to IBM DASD rules.  That means no more BS standards wars between knuckle-draggers in the storage world who continue to mitigate the heterogeneous interoperability and manageability of distributed systems storage using proprietary lock in technologies designed as much to lock in the consumer and lock out the competition as to deliver any real value.  That has got to be worth something.

You may be asking, is Toigo out of his mind?  Can’t he remember what a pain in the tuccus IBM was during its hey day as big iron maven?  The answer is, I remember all too well IBMs bad behavior.  I remember kicking my account rep out of my office and barring him from ever setting foot on my campus again.  Do I want to return to this state of affairs?  Nope.

There are some holes in the everything virtualized on the MF, of course.  For one, what do we do with all the Microsoft servers.  There is no Redmond-sanctioned approach to my knowledge for virtualizing Microsoft SQL Server or Exchange Server in a mainframe partition.  The OS is quite particular about resource handling and doesn’t seem to want to work and play well inside an emulation environment.  This explains in part why MS wares appear to run so much better in a MS Hyper-V environment on Server 2008 than in a VMware environment.

I know we can move a boatload of UNIX and Linux servers into mainframe LPARs.  When I ask about Microsoft running in an LPAR, I get a wink and a nod.  It can be done.  In fact, there are several approaches for doing so, though none are documented as nearly as I can tell at IBM and Redmond doesn’t seem to be pursuing the approach with any sort of enthusiasm.  That is a gotcha for mainframe centric server consolidation in my book.

I have been talking to a lot of folks about this and would like any input anyone else might want to provide.  

{ 3 comments… read them below or add one }

zax218 March 18, 2008 at 9:13 pm

First prblem with the Z10, initial capital investment of at least 1 million in server hardware. a This does not include storage, OS licensing, and SW licensing.

Most companies have deployed mainly web servers, app servers, file servers, and print servers on ESX. Why? Because they are the low hanging fruit with low utilization numbers. changes in virtualization technology and processor technology are allowing for the virtualization of servers with high I/O loads. Raw Device mapping and NPIV have really changed the game.

Combine Blade technology, ESX embedded, and Distributed power management, and you have a powerful solution to reduce power consumption in your data center. Consider it may take a year to fully utilize a Z10. Does the system have power management to reduce consumption of power while the system is not fully utilized?

While it does take highly skilled personell to deploy and manage ESX, it also takes highly skilled individuals to manage a mainframe environment. Your average level I windows Administrator will not be able to participate in the provisioning process.

Show me data where MS Wares run better on Hyper-V over ESX? The product is not even commercially available. Since SP2 in windows 2003 both solutions are about even. As pointed out, main stream windows applications are not even supported in this environment(Z10). Yes, you can make it work..I can also run OSX on my Intel laptop….that doesn’t make it an option for the enterprise.

A few other points:

Storage options are significantly cheaper and more widely available for ESX.

Investing in a 1500 machine system requires a significant capital investment in hardware and migration time. Because of the high price of the system, enterprises are forced to fully utilize the hardware ASAP because the maint clock starts on day one for the hardware. We are not talking 20% maint on 20K. Maint is a significant Opex for a system this large, which will drive the migration strategy toward an expensive PS engagement.

What happens in 4 years when you need to upgrade the hardware? Another PS engagement.

DR and backup options are more widely available for the ESX platform.

Operational efficiencies on ESX should be included in this discussion. Server build time, backup/restore times, snapshot and recovery of servers…etc. All from a single console performed by a level I Administrator.

One value of ESX is overcommitment of resources. I do not think this is available in the mainframe world. I have a farm that averages 35-40 VM’s on an 8 core 24GB RAM system. Your numbers per VM are most likely not taking into account this concept.

At the end of the day you have a commodity based solution that can be scaled on demand vs a big Iron solution with limited support from existing vendors.

This is not much different than HP’s push of the Superdome. If you are a big unix shop, this is a an option for you…everyone else. Look elsewhere.

Administrator March 20, 2008 at 10:44 am

Hey look, I was reporting, not judging.

You make some good points and some not so good points.

Storage costs and ecosystems around VMware can be debated readily I think. Given the much better tools for capacity and data management in the mainframe world, your total cost of ownership for storage is far less than your cost for distributed storage: look at the difference in allocation and utilization efficiency between mainframe and x86 storage published by Horison. 70% of optimum, versus a pawltry 17% for distributed systems storage. That’s what DFSMS and DFHSM buy you on the mainframe.

I did say that $1 million entry cost on Z10 was painfull to those who did not already have a mainframe investment. That said, I look at the price of the blade servers you described — or even beige servers — which run up to $20K for those I might trust to host multiple virtual machines and generously allocate 20 VMs per server. Add in the licensing costs for VMware and the need (as you point out) to add in third party software to do DR failover over distance, and pretty soon you are way over the $1M for 1500 x86 virtualized servers offered by IBM.

TCO aside, a big issue is who’s bitch you want to be. VMware’s, Microsoft’s, or IBM’s. I am not about to give an opinion on that one.

zax218 March 21, 2008 at 3:42 pm

I made a purchase last year that was slightly under 1mil (823K Cap / 161 Op).

I won’t get into vendor specifics, however all the vendors used were major enterprise players. Everything has 3 years support included.

This purchase includes all licenses, storage, and server hardware to host a minimum of 1,000 windows server virtual machines. It also includes licenses for DR software and other misc utilities in the environment.

My number of 1,000 VMS does not take into account the use of overcommitment of resources which can grant huge savings. Each VM has 768MB of RAM and 20GB of hard drive space. 768MB of RAM runs well on Windows/ESX because of the memory synergy created with Transparent Page Sharing (TPS).

Add OS licensing to your Z10, and you are almost at 1.5M already (estimate $300 for a windows license, virtualization has a better licensing model when using Windows datacenter edition). Start adding in storage and other licensing (AV, Monitoring software, etc) and you are approaching 2M with a capacity of 1500VMS.

My point is that the price is just about a wash when comparing the two. Back to my original post; “If you are a big unix shop, this is a an option for you…everyone else. Look elsewhere.” I don’t see much overlap between this solution and a server virtualization solution on the x86 platform.

Who’s bitch do I want to be? Probably VMWARE. They don’t want/try to sell me hardware and storage and I am free to pick any vendor I choose to run my virtual machines on. While virtualization software is becoming a commodity (see the plethora of good hypervisors that are available), hardware is currently a commodity that I can drive the price down on.

*I can provide you all the numbers you want in relation to this purchase.

Previous post:

Next post: