Other virtualization techniques

---

LPARs in a virtualized server environment

Logical partition or logical partitioning (LPAR) is another way to achieve a high degree of isolation between applications sharing the same physical server. LPAR is a feature that allows multiple instances of operating systems to run concurrently on the same hardware. A high degree of isolation can be achieved because each LPAR has its own installed instance of the operating system. If there is a problem that causes the operating system to be affected, the worse case effect is only on the software and applications installed on one affected partition.

Early versions of the zSeries mainframes were first to introduce the concept of LPARs. Today LPARS are also supported on the pSeries and iSeries. All new models of the pSeries servers support LPAR (p630, p65x, p670 and p690).

LPARs have many benefits in a virtualized environment. The LPARs can be very beneficial to a quality assurance (QA) test environment. A QA test environment is usually a low transaction volume environment, sometimes under constantly changing environment settings at the operating system, middleware, and application level, due to debugging, and coexistence, integration, and system testing activities. The benefits of LPARs to this type of environment include:

• By definition the test environment runs high-risk scenarios, as untested application code can be volatile. The isolation provided by having a separate operating system instance is very desirable. In system test scenarios the operating system settings may vary from test case to test case or day to day.

• System management is facilitated by the fact that an LPAR definition can be saved to disk and only loaded when needed. This means that when tests are not running on an LPAR, that LPAR's resources can be made available for another LPAR. When the LPAR is needed for testing it can be brought up again with proper scheduling, but without the need to re-create any environment settings. For example, a 4-way p630 server can have up to four concurrently running LPARs, however far more than four can be defined and saved for occasional use.

Another benefit of LPARs is better scalability for servers with eight or more CPUs. For example, a 32-way pSeries 690 will generally provide greater throughput for a WebSphere application if it's divided into four 8-way LPARs, and the same for an 8-way xSeries if it's divided into two 4-way LPARs.

Virtual machines

Virtual machines are similar to LPARs in that they create a high degree of isolation between workloads running on the same physical machine. The VM operating system on the IBM zSeries mainframe has provided this option for decades. Recently this has become an option on other platforms, in particular the Intel-based xSeries platform.

VMW is one virtual machine technology that supports the xSeries platform. Virtual machine emulators are more flexible, but can be less resilient, as the overall stability of the system is limited to that of the host operating system. The flexibility comes with the number of instances and ease of management of the virtual machine. The number of CPUs does not limit the number of instances that can be supported. For example, a two-way machine can have up to 16 virtual machines. This type of configuration is ideal for test environments where transaction volume testing is not a concern.

Advanced pSeries features

In support of on-demand computing, IBM introduced advanced functions for the pSeries that can significantly improve server virtualization. Two of these key functions are dynamic LPAR (DLPAR) and Capacity Upgrade on Demand (CuOD).

DLPAR means that the resources assigned to LPARs can be reassigned or redistributed amongst active LPARs without the need to stop the execution of the LPARs. For example, a CPU can be reassigned from one LPAR to another LPAR, without restarting the operating system on either LPAR. DLPARs are ideal for sharing low and high priority workloads on the same server. When the high priority workload requires more resources to meet agreed to service levels, resources can be dynamically shifted from the low-priority LPAR to the high priority LPAR. For more information about the details of the LPAR function see: www.ibm.com/servers/eserver/pseries/hardware/whitepapers/dlpar.html

CuOD allows pSeries customers to buy only the number of CPUs required to achieve their SLAs today. As workload throughput requirements increase, customers can rent or purchase more CPUs. The additional CPUs are activated without the need to physically add CPUs. The CPUs can be dynamically assigned to an LPARed environment. For more information on Capacity Upgrade on Demand see www.ibm.com/servers/eserver/pseries/cuod/

---