Making the Case for System z10 Server Consolidation
Published: March 25, 2008
by Timothy Prickett Morgan
Several weeks ago at the launch events for the System z10 mainframe in New York, Tokyo, and Mumbai, IBM's top brass in the Systems and Technology Group and Software Group did not spend a lot of time talking about the feeds and speeds of the technology inside of the new z10 box. While there was some talk that the machine could support hundreds of millions of users, based on a benchmark test in China, a lot of the conversation was about virtualization and server consolidation.
The reason, of course, is that a lot of the revenue that IBM is getting from mainframe shops these days--at least the incremental revenue that has kept overall mainframe sales from declining--is coming from consolidation of Windows and Linux workloads on X86 and X64 servers and of Unix workloads on RISC servers onto Integrated Facility for Linux engines on mainframes. IBM can't sell the idea of consolidating mainframe workloads because this has already long since happened at the mainframe shops of the world. The cheapening of telecommunications, the recession in the late 1980s and then early 1990s, and the transition to necessarily cheaper mainframes engendered by competition with RISC/Unix machines, and the relative sophistication of mainframe operating system virtualization is what caused the mainframe revenue stream to collapse by half in the early 1990s, and it contracted further until $4 billion a year in mainframe sales for the past couple of years is cause for celebration.
The server consolidation sales pitch is one that today's mainframe shops will listen to, post dot-com build out, and it is one that the IT managers responsible for mainframes can use to prop up their own personal empires as they are, like IBM's mainframe business, surrounded by what appears to be cheaper X64 alternatives that are growing in sophistication and that sport a much larger application portfolio. The empire rebuilding that some VPDPs and CIOs at mainframe shops are engaged in today is an important component of the success of mainframe consolidation products, because even with very strong economical arguments, companies do not generally redeploy applications on new platforms. This is why mainframe applications have persisted for decades, even after waves of data center and then mainframe footprint consolidation in the 1980s and 1990s, and then attempts by Unix and then Windows and Linux platform sellers to take workloads off the mainframe. There has to be something more than compelling economics, but compelling economics is the beginning of the process.
To be sure, a lot of companies have moved off the mainframe--enough that there are only somewhere between 10,000 footprints (IBM's number) and 12,000 footprints (IDC's number) out there in the world today. But all footprints are not the same size, now are they? If IBM can't add footprints for the mainframe easily, the excellent virtualization characteristics of the z/VM operating system, which allows guest operating systems to run inside it (including Linux and soon Solaris), and the ubiquity and corporate acceptance of Linux for certain kinds of workloads means that IBM can sell bigger mainframes to customers who simply cannot add another dozen racks of X64 iron to their data centers.
While Gartner and IDC may be projecting a world where there is 45 million servers running in the world by 2010--there were 5 million in 1996--very few companies want to pay for managing the physical server sprawl. And considering that many of those physical servers will be running a handful of virtual machines at a time and will probably have tens of images that can be deployed to them as workloads change throughout the day, month, and year, what we are talking about is potentially between 1 billion and 2 billion virtual server images sitting on a jukeboxes on the networks of the world that have to be managed; at the very least, several hundred million images. We are coping with maybe 30 million or so images today, and the company bean counters are screaming that companies are spending way too much money on system administrators. We are talking about an order of magnitude in sprawl increase. (That is my estimate, not IBM's.)
And that is why IBM is very keen on talking up the 64-way z10 and the hundreds to thousands of Linux server images it can all host on a single, virtualized machine.
In his presentation two weeks ago, Steve Mills, senior vice president and the group executive in charge of IBM's Software Group, did the launch of the z10 in New York. In his presentation, he talked about how the z10 mainframe could, running at near peak performance, replace as many as 1,500 underutilized X86 servers, and do so with an 85 percent reduction in floor space, an 85 percent smaller energy bill, and with a 30 to 1 consolidation of software licenses, which are generally calculated on a per socket or per core basis.
In another comparison, Mills took a bunch of servers comprised of 760 computing elements (cores in the X64 world and CPs in the mainframe world) that were unvirtualized. And then the workloads were virtualized using an X64-based hypervisor (presumably VMware's ESX Server) and they could be consolidated down to X64 servers with a total of 304 cores. While this consolidation boosted the utilization of hardware and would, for a new project at least, mean spending less money on hardware, the software bill, thanks to the hypervisor and related management tools, would mean that this hypothetical shop could only reduce the overall cost--which includes hardware, power, personnel, software, and maintenance costs for both hardware and software over an unspecified time--by about 26 percent. (In plain English, if it costs $100 to do the unvirtualized servers, it costs $74 to do the virtualized servers. Now, in this scenario, you are reducing the processor core count by a factor of 2.5, and presumably driving up utilization by a factor of 2.5.) So now roll in a System z10. IBM says that it can handle the same workload, but do it on a mere 26 of the z10 engines (that is six and a half of the quad-core z6 processors running at 4.4 GHz) and do so at 20 percent of the cost of the original X64 server setup. In this scenario, the money shifts away from power and people and other people's software and toward IBM's mainframe hardware and z/VM software, of course.
But IBM's sales pitch for z10 server consolidation is not just about a static comparison like this. It is about being able to easily deploy new operating system images and scale the server up, almost instantly, to meet the kind of unexpected processing demands that can happen thanks to the Internet. To help make its case for the System z10 as a Linux consolidation platform, IBM invited Buzz Woeckener, manager of zLinux environments at insurance carrier Nationwide to tell the story of how that company moved from X86-Linux machines for its Website and related insurance brokerage applications to Linux images running on a pair or mirrored z9 servers.
Nationwide, according to Woeckener, had over 5,000 servers of many favors and fancies running in its data center to support the Website. The insurance company was projecting that it was going to run out of power and cooling capacity in its data center--which is a big tier four data center, making it one of the largest ones in the world. The company was anticipating that it would have to spend tens of millions of dollars to build out its data center, and the idea to shift Linux images to a mainframe was originally proposed as a stop-gap measure, just putting off the inevitable data center build out. As part of the sales pitch, Woeckener hoped to speed up the time it took to provision a new server, which could take from weeks to months with physical X64 servers; he also hoped to be able to shift more performance underneath certain partitions, which need more oomph on Friday afternoons. The company was also looking to add some disaster recovery capability to the Linux and Unix workloads supporting the Nationwide Website and its applications.
Within 90 days, Woeckener had a prototype set of Linux partitions running on its mainframes, which supported other applications, and after a successful test, Nationwide made the decision to set up two data centers linked to each other and backing each other up through disk-based mirroring. The production z990 mainframe used by Nationwide for its Website has 18 processors activated and it runs four logical partitions, backed by 240 GB of main memory and eight I/O cards; a second machine, which is used for application development and testing as well as a hot backup site for the production machine, is a z990 with 17 cores activated running five LPARs; it is equipped with 325 GB of main memory and has a dozen I/O cards. The machines run z/VM 5.3 and have a total of 484 Linux instances, Novell's SUSE Linux Enterprise Server 9, that run IBM's WebSphere application servers and support a total of 18 different Java-based Web applications. The setup supports in excess of 100,000 active users--clients and brokers who access claims--each day, and was deployed in four months.
Woeckener says that because of its move to Linux on the mainframe, Nationwide expects to save $15 million over three years. The monthly costs of Web hosting have been reduced by 50 percent, and so has hardware and software support costs. The z990 mainframes take up only 20 percent of the space of the gear they replaced, software costs dropped a lot, and processors are not running at somewhere between 85 percent and 90 percent. Server image provisioning happens in days, not months, and a number of times, the company has been able to add capacity quickly and painlessly--such as when Nationwide ran a Super Bowl ad. This was something that would have been a nightmare using X64 servers. More importantly, Nationwide now expects it will not have to build out its data center until 2010.
"This has been a real good ride for Nationwide," said Woeckener. "This has delivered everything we needed it to."
Which is cool. Because Nationwide can now lower my car, co-op, and business insurance, right?
IBM Launches 64-Way z10 Enterprise Class Mainframe Behemoth
Thinking Inside the Box
A Mainframe Renaissance
IBM Touts the Power Efficiency of Mainframe Linux
IBM Readies Quad-Core z6 Chip for Mainframe Iron
The IBM Mainframe Base: Alive and Kicking
The X Factor: One Socket to Rule Them All
HP, Intel, and Oracle Gang Up on IBM Mainframes
It's MIPS Again, at Least for Hardware Pricing
A New Mainframe Deal or Pollux as Usual?