Virtualizing Workloads with SUSE Linux Enterprise Server for System z Eliminates Server Sprawl
Written by Meike Chabowski
Virtualization in distributed computing environments is wildly popular—with good reason. It's dramatically reducing the numbers of x86-based physical servers in IT environments everywhere. However, for many organizations with high-volume transaction and data-processing needs, there’s a smarter way to reduce physical server numbers and total cost of ownership while providing the highest levels of scalability, availability, energy efficiency and footprint reduction. The smarter way is virtualizing workloads using the z/VM hypervisor and SUSE Linux Enterprise Server for System z.
But Mainframes Are So 1960s!
Maybe, but so is virtualization. Virtual machine technologies were first developed by IBM for its System/360 mainframe in 1964.1 What’s more, mainframes have never gone out of favor in many organizations. For banking, finance, health care, insurance, utilities, government, and a multitude of other public and private enterprises that need reliable and powerful data-processing capabilities, mainframes have been continuously on the job for decades. More than 95 percent of Fortune 500 companies use mainframes.
Today, in many organizations, mainframes are taking on more of the computing load than ever before, largely due to the benefits of SUSE Linux Enterprise Server for System z. Those benefits include cost reduction, access to hundreds of Linux applications and a global community of Linux experts, compatibility with other systems running Linux, and Novell support.
And there are other compelling benefits.
Server Sprawl Is Obliterated
Physical server numbers can be reduced by factors of 100-200:1. This compares to around a 10:1 reduction through virtualization in a homogeneous distributed computing environment.2 When it comes to reducing space requirements, the mainframe wins hands down.
Today, in many organizations, mainframes are taking on more of the computing load than ever before, largely due to the benefits of SUSE Linux Enterprise Server for System z.
CPU Use Goes Way Up
On average, x86-64 platform CPU use rates of 10 to 15 percent are considered normal. With virtualization technologies, you can take those numbers up between 25 to 50 percent. But even in rare “high-use” cases, you are still stuck with the scale-out approach to workload growth that drains power and hogs real estate. System z, on the other hand, normally operates at anywhere from 85 to 100 percent use and takes about as much floor space as a big refrigerator.
Licensing Costs Come Down
Consolidate software stacks and dramatically reduce CPU-based software licensing costs with the Integrated Facility for Linux (IFL), the specialty engine for Linux workloads on System z. An IBM System z10 Enterprise Class mainframe can reduce per-core licensing costs up to 30:1 compared to x86 environments4. This is because a single System z server can run multiple Linux applications on a single processor engine, and enterprise Linux software is usually priced on a per-engine basis.
Support Staff Requirements Level Off
Consolidating multiple servers onto a single System z mainframe running multiple virtual Linux servers can reduce the labor required for system management and maintenance. The centralized system management and autonomic computing features of SUSE Linux Enterprise Server for System z can also help cut down on the errors and workload-balancing tasks that can devour IT staff time.
Electric Bills Drop Like Rocks
Power and cooling cost reductions can exceed 80 percent. Distributed computing can’t compete when it comes to cost per kilowatt. In fact, according to one analyst, the IBM System z platform can be configured to require 1/12th the electricity of a distributed server farm with equivalent processor capability.5
Data Center Footprints Shrink
Expect data center floor space reductions of up to 25:1. The System z10 Business Class (z10 BC) mainframe has the capacity of up to 232 x 86 servers with 83 percent smaller footprint.6 Basically, the difference comes down to one big box versus a full room of server racks (and all of the associated cables and switches).
Server Provisioning Accelerates
Provisioning an x86-based server can take weeks or months. Provisioning a server partition on a System z machine can be accomplished in just a few hours.
Security Becomes Less of an Issue
Securing virtual servers at the hypervisor level is an ongoing concern in any distributed computing environment. However, with the level of security inherent in the System z platform, it’s not something that typically worries mainframers.
IT Environment Simplifies
Distributed server solutions can entail three to four times the number of servers that a mainframe production environment requires just to address workloads such as test/development/QA and D/R.7 All these workload requirements can reside within a single System z mainframe using existing virtualization techniques like z/VM and PR/SM LPAR, which are common to most System z configurations.
TCO Is a Pleasant Surprise
Combining the benefits of System z servers with those of SUSE Linux Enterprise Server for System z creates an entirely new TCO value proposition. This is due in part to Integrated Facility for Linux (IFL), which costs significantly less than typical central processors. Linux on mainframes gives organizations the opportunity to add more power to existing "Big Iron" infrastructure at a fraction of the cost of a non-Linux deployment. In a high-volume, high-transaction environment, TCO reductions range from 30 to 50 percent compared to distributed computing. Of course start-up costs for non-mainframe shops might seem high at first sight, but if planned right, substantial savings can kick in as the system grows in use.
Here Comes the Cloud
Everyone has their own take on what the cloud is and what it can do for their organization. But regardless of what people hope to accomplish with cloud computing, the common denominators are agility and paying for what you need. On-demand compute power makes the cloud possible, and that’s been the foundation of SUSE Linux Enterprise Server for System z and the z/VM hypervisor from the start. The Linux/System z combination is a powerplant that is highly scalable, available and reliable by design. It’s a workhorse with costs that are fixed, manageable and predictable. What’s more, IBM mainframes have a mean time between failure (MTBF) that isn’t just measured in years, but decades.
So, SUSE Linux Enterprise Server for System z is a natural platform for delivering the next generation of utility computing services, whether in private or public cloud scenarios. The core requirements for these applications—massive resources coupled with intelligent workload management (IWM) features capable of supporting extremely dynamic workloads—map perfectly to the features of SUSE Linux Enterprise Server on System z.
It’s no wonder Linux on the mainframe is the foundation of IBM’s cloud strategy.
Innovations in the SUSE Linux Enterprise Server Distribution
It has been ten years since Linux premiered on the mainframe. SUSE Linux Enterprise Server for System z has been available since the outset. Year after year, it has gained popularity—retaining 80 percent of the zLinux market. And Novell keeps adding functionality.
A critical tool for adding functionalities and innovation—a tool that is unique to Novell—is the Internal Build Service (formerly known as AutoBuild). It enables Novell to create a feature or capability once—or fix a problem once—and release it on all Linux platforms the company supports.
Key innovations include the SUSE Linux Enterprise Server for System z Starter System. It makes installing Linux on mainframes a lot more user friendly by letting you transfer images to z/VM using FTP. Then, using z/VM commands and utilities, you can create a Linux guest that is a full-blown installation server that can be used to create other Linux guests under z/VM as necessary. The image can include a complete SUSE Linux Enterprise Server OS configured with FTP, HTTP, Samba and other servers. All you need to provide is an IP address and some disk space.
In addition, Novell is working with partner IBM to bring new workloads to the mainframe. Customers can now run .NET based applications on their mainframes thanks to the availability of SUSE Linux Enterprise Mono Extension for System z.
Nationwide Insurance IT management was looking for a way to dramatically reduce the total cost of ownership of their distributed server and Web hosting environment. They found it with SUSE Linux Enterprise Server for System z—“it” being reduced complexity, shorter provisioning times, fewer staff requirements (no physical installation to perform), a smaller carbon footprint and much reduced floor space requirements.
The following performance basics are courtesy of Rick Barlow, Senior z/VM Systems Programmer, Nationwide Insurance. They are excerpted from his presentation at the SHARE IBM user group conference on March 17, 2010.
Basic Metrics to Watch: z/VM
- System z runs fine at 100 percent, but Linux workload is much more demanding than traditional mainframe workloads.
- Significant impact of memory over-commit
- May need to keep peak periods at 85-90 percent
- Many Linux guests have huge working set sizes and many don't go idle
- Keep memory over-commit less than 2:1: (ratio of combined working set sizes to real memory available)
- z/VM has no problem with high page rates: Keep expanded storage for high-speed page buffer
- Guests may not be tolerant
- Allocate enough VM page space for twice the total of the working set of expected guests; Use CP QUERY ALLOC PAGE to monitor and keep page space less than 50 percent full
Basic Metrics to Watch: Linux Guests
- Don't wake guests to ask: Choose performance tools that understand that Linux is running on z/VM
- Pick one tool: Multiple monitoring tools adds a lot of overhead
- CPU measured inside guest is not very meaningful
- Avoid TOP—significant overhead: Use vmstat or nmon
- Don't over provision. Large virtual storage sizes drive up z/VM paging
- Use a swap hierarchy—it is not a problem for Linux to do some swapping
- Show all snapshot of memory/swap
- Avoid multiple I/O caching in DB2
- Default Linux memory management may not be optimal
- Prevent Linux from paging. z/VM paging is much more efficient.
- Show Linux page-in/page-out
- Look at guest CPU demand from z/VM
- Watch for excessive paging on behalf of a guest this may indicate inefficient memory usage or excessive virtual storage allocation
- Watch for guests with poor I/O response: System z handles high I/O rates fine but bottlenecks can occur
- Watch for percent of active time that guests spend in various queues
Basic Metrics to Watch: Linux Guests Internal Performance
- Tools to analyze guests functions vary greatly—pick the right one
- Application developers debugging skills may be limited: They might be accustomed to working with excessive capacity, but not accustomed to shared environment
Ideas that May Help
- Utilize Cryptographic hardware. This dramatically improves SSL calls for secure Web pages
- Minimize external network hops
- Reduce NTP frequency
- Minimize or stagger cron scheduling
Linux on System z Is Extremely Attractive
Mainframes will continue to play a vital role in the enterprise for the foreseeable future. It’s not an overstatement to say that System z is the ultimate virtualization resource—a massive data center consolidation platform capable of up to 60 logical partitions and the addition of hundreds of Linux virtual servers under the direction of z/VM technology. All in a single rack.
Moreover, SUSE Linux Enterprise Server for System z is a cost-effective open platform for development. In fact, mainframe growth is largely being driven by Linux developers using Linux with Apache, MySQL and PHP (the LAMP stack).
Through the years, IBM has consistently improved the price/performance ratio of System z mainframes, and IFLs are lowering licensing requirements and costs for SUSE Linux Enterprise Server in a big way. Many companies that conduct cost-per-transaction analyses find that SUSE Linux Enterprise Server for System z is unbeatable—in some cases as much as 50 percent lower than distributed environments. And, of course, Linux on mainframe environments are easier on the planet. In that context, upfront mainframe costs don’t look so bad after all.
- http://www-03.ibm.com/systems/z/advantages/energy/index.html5 http://www-03.ibm.com/systems/z/os/linux/about/#foot6 ftp://public.dhe.ibm.com/common/ssi/sa/wh/n/zsw03162usen/ZSW03162USEN.PDF
- IBM System z: Platform Star for Linux and Open Source Software, Ptak, Noel & Associates, as cited at http://www-03.ibm.com/systems/z/advantages/energy/index.html
- IBM System z, the Smarter Mainframe, IBM white paper (03/2010),