Partitioning: Your Mainframe becomes a Hotel for OS 'Guests'

Computerworld - Savvy IT pros have been tinkering with mainframe partitioning for years. But with the sudden proliferation of single-task servers, especially for the Internet, partitioning big iron to consolidate those operations is more important than ever.

But partitioning can be a challenging move, too. The first step is to identify the business requirements of your workloads. Determine end-user expectations for application response time and identify all service-level agreements for your operations.

With that in mind, your next step is to simply count: How many discrete operating environments -- or "guests," in IBM-speak -- will you need? That is, how many instances of z/OS, legacy OS/390, Linux or other operating systems will the hardware be running?

Armed with the performance requirements and the number of guests you'll be hosting on your mainframe, you can choose which of the following approaches is best for carving up your mainframe's resources:

1. The Major Hotel
If you plan to populate your mainframe with a lot of guests, the Virtual Machine (VM) approach is right for you. An operating system itself, VM is the go-between for operating system guests and the hardware resources.

Just as when you book a stay in a major hotel and choose a suite, a penthouse or a regular room, you create defined resources for every guest operating system on the hardware through a management program accessed with the VM console. Each VM guest includes the amount of I/O, processor and memory assigned to it. The VM approach can handle thousands of partitions, though realistically the numbers range from a few dozen to a few hundred.

These aren't dedicated physical resources, but virtual ones. VM sorts out which guest gets how much of what resource and when, depending on the business rules and available capacity.

Randy Lengyel, senior vice president of MIS at Wisconsin Physicians Service Insurance Corp. (WPS) in Madison, says that once you set up your partition template in VM, "it just takes two or three minutes to set up a server."

Earlier this year, WPS began a Linux server consolidation effort that it intends to complete by next month. Already six of its 30 Linux boxes destined to be partitioned have been redeployed and the Internet applications they ran are now housed in their own z/900 partitions.

2. The Bed & Breakfast
A logical partition (LPar), lets you carve up a system's physical resources and assign them to a given operating environment. Memory must be allocated to logical partitions, but it can be dynamically reconfigured when needed. Processor resources and I/O can either be dedicated or shared among logical partitions. With this hardware-based approach you can have up to 15 LPars on a mainframe.

That limitation of 15 partitions matters to some users. "We chose VM over LPar because it gives us more flexibility to add partitions," says Jim Hwong, director of enterprise network systems at WPS.

LPar does have some advantages. David Boyes, chief technology officer at Sine Nomine Associates Corp., a consultancy in Ashburn, Va., says, "With LPar you get really separate machines on the same mainframe." This is ideal for workloads that have specific performance and capacity requirements, he adds.

But Noyes cautions those who go the LPar route because of its complexity. Just charting out the various I/O path options can be daunting to even a veteran system manager.

"It's complex. LPars are static and need to get partitions set right," he says. "If they're not done right and you need to reconfigure, you have to stop the machine."

The tool of choice for establishing LPars is IBM's Intelligent Resource Director (IRD). After feeding it the business goals and workload criteria, the IRD identifies ideal partition configurations. During operations, IRD assures that each partition gets the resources necessary and that the overall system is running at optimal performance. For example, when a processor goes into a wait state, it signals that it's available for other tasks, which IRD can then send its way.

3. The Hybrid
You can also run VM in an LPar, giving you the best of both worlds, albeit a much more complex one. Plus, you will need to hire people versed in both partition schemes.

LPar treats VM as a single operating environment, taking one of its 15 partitions, while VM manages its guests per normal.

One clear benefit of partitioning is better reliability. "The Intel processors on Linux servers have a mean time to failure of about six years," Lengyel estimates. "The CPUs on the z/900s have 30 years. Now that's uptime."

Read more about Networking in Computerworld's Networking Topic Center.