This section briefly explains virtualization components and features and provides quick references to the virtualization environment. Readers familiar with virtualization terminology and concepts can skim through this section.
|
Logical partition (LPAR) or partition |
A logical partition is an isolated computing domain with its own resources (processor, memory, and I/O interfaces) and operating system instance. Supported operating systems include the AIX, Linux (RedHat, SLES), and IBM i operating systems. Each LPAR can run a different type, version or level of an operating system. For example, one LPAR can run AIX 5L™ v5.2, a second LPAR can run AIX 5L v5.3 TL06, a third LPAR can run AIX 6, and a fourth can run the Linux operating system.
In addition to processor and memory resources, each LPAR needs its own root disk, network interface, and storage. There are ways to simplify and share network and storage adapters using virtual I/O, as explained in more detail later.
There are two types of LPARs:
- Shared processor (which uses the IBM Micro-Partitioning feature).
|
|
Dynamic logical partition (DLPAR) or dynamic reconfiguration |
The DLPAR facility lets you alter the resources of a partition at run time, without restarting the operating system. Examples of items that you can alter are the number of processors for dedicated partitions; the number of virtual processors and entitled capacity for shared processor partitions; and the number of virtual I/O adapter slots and the amount of physical memory for either type of partition. The DLPAR facility enhances resource utilization by allocating resources where they are needed most.
You can access the facility manually through the Hardware Management Console 1 (HMC), or you can automate the access by using a workload management tool. The DLPAR facility is essential for workload management tools such as IBM Enterprise Workload Manager™ (EWLM). |
|
POWER Hypervisor (PHYP) |
PHYP acts as the abstraction layer between the system hardware and the LPARs, enabling multiple operating systems to run on POWER processor-based systems. PHYP is a key component of the IBM virtualization technology that enables Micro-Partitioning, shared processor pools, dynamic LPAR, virtual I/O, and virtual LANs. Among the many tasks that PHYP performs is saving and restoring all processor-state information during LPAR context switching. |
|
Dedicated processor partition |
A dedicated processor partition has one or more assigned processors exclusively reserved for it by PHYP. (You can assign processors in increments of one.) While this partition is active, other processors cannot use the idle processor capacity. PHYP uses the same physical cores to schedule partitions to benefit from a warm cache.
Dedicated processor partitions can improve LPAR throughput through processor and memory affinity and help ensure maximum processor cache hierarchy performance.
This feature applies to the POWER5™ 2 family of processors. |
|
Shared processor partition |
You can assign processor capacity to a shared processor partition in increments of 1/100th or 1% of a physical processor. However, each partition requires a minimum of 1/10th or 10% of physical processor capacity. As a result, there is a maximum of 10 partitions per physical processor.
Shared processor partitions require the IBM Advanced Power Virtualization (APV)3 feature and use the IBM Micro-Partitioning feature. |
|
Entitled capacity |
Entitled capacity refers to processor capacity that is assigned to a shared processor partition.
This capacity is a guaranteed amount of processing capacity. PHYP slices up physical processor time to manage fractional processor allocation. |
|
Capped shared processor partition |
Entitled capacity is the hard limit for a capped shared processor partition (that is, it cannot exceed its entitled capacity when demand is high, even if idle processing capacity is available in a shared pool). Capped shared processor partitions enable you to better manage the shared-pool processor resources. |
|
Uncapped shared processor partition |
An uncapped shared processor partition can use not only its entitled capacity but also available idle processing capacity in the shared pool. Idle shared-pool processor capacity is immediately yielded to meet workload demand at peak times when a partition needs more than its entitled capacity. Uncapped partitions are very flexible and useful for unpredictable workloads.
When there is more than one uncapped shared processor partition, shared-pool processor capacity is assigned to a shared processor partition based on its assigned uncapped weight. |
|
Uncapped weight |
The uncapped weight is a number that specifies the portion (weighted average) of shared-pool idle processor capacity to be assigned to a shared processor partition when there is more than one overburdened shared processor partition.
The uncapped weight is a number between 0 and 255; the default uncapped weight is 128. The higher the uncapped weight, the more idle shared-pool processor resources are granted. You can set the uncapped weights for various shared processor partitions according to partition workload priority, setting the highest uncapped weight for the highest-priority workload. |
|
Virtual processor |
This concept is relevant to shared processor partitioning only. A virtual processor is the entity to which the operating system dispatches application threads (or processes). You can view the virtual processor as a processor in a traditional nonpartitioned environment.
For example, assume that a shared processor partition has 1.50 entitled capacity with two virtual processors running AIX 5L. In this case, the AIX kernel scheduler acts as if it has two real processors. Assuming simultaneous multi-threading (SMT) is turned OFF, the operating system schedules a maximum of two threads at a time.
You can consider the virtual processor to be virtual core. The number of virtual processors is unaffected by SMT. |
|
Virtual I/O (VIO) |
Virtual I/O is a broad term that refers to a set of storage and network virtualization features:
- shared Ethernet adapter (SEA)
Virtual Ethernet. Without requiring additional hardware or external cables, a virtual LAN (VLAN) facilitates high-speed virtual Ethernet communication paths among multiple partitions within a physical system that run AIX, Linux, and other operating systems. You can dynamically create virtual Ethernet segments and restrict access to a VLAN segment to meet security or traffic segregation requirements. A virtual Ethernet has the same characteristics as a high-bandwidth, physical Ethernet network and supports multiple networking protocols, such as IPv4, IPv6, and ICMP. PHYP provides this feature.
Virtual I/O server (VIOS). This is a special-purpose partition that provides virtual I/O resources to client partitions. The VIOS owns physical adapters. You can share a physical adapter by assigning it to multiple client partitions, which minimizes the number of physical adapters that you require for individual clients. Thus, the VIOS can reduce costs by eliminating the need for dedicated network and disk adapters. The VIOS provides two major functions:
- A virtual Ethernet bridge. The shared Ethernet adapter (SEA) hosted in the VIOS acts as a layer-2 bridge between the internal virtual and external physical networks. The SEA enables partitions to communicate outside the system without having to dedicate a physical I/O slot and a physical network adapter to a client partition.
- A virtual SCSI adapter. With this virtual interface, physical storage (in cases of device backing) and logical volumes (in cases of logical volume backing) that you create and export from a virtual I/O partition are shown at the client partition as a SCSI disk.
VIOS requires the APV4 feature. |
|
Free-pool capacity |
Free-pool capacity refers to the amount of free processing capacity available for use by uncapped shared processor partitions. You can calculate this amount by adding the unassigned processing capacity (that is, it is not guaranteed to any LPAR using processor entitlement or by virtue of a dedicated LPAR) to the sum of unused entitlements for all the shared processor partitions active in a system. |
|
Live Partition Mobility (LPM) |
Live Partition Mobility is available starting with POWER6 processor-based systems. LPM is designed to enable migration of an LPAR from one system to another compatible system. LPM uses a simple and automated procedure that transfers state and configuration information from the source server to the destination server without disrupting the hosted operating system or applications. This feature requires a VIOS. |
|
Workload partition (WPAR) |
A workload partition is a new feature in AIX 6.1 that creates multiple AIX virtual partitions based on a single global AIX install. One can create a System WPAR or an Application WPAR. DB2 product installation is currently supported only on the system WPAR. WPAR mobility enables partitions to be moved from one physical system to another. Refer to the AIX and DB2 documentation for minimum AIX and DB2 levels to install the DB2 product within a WPAR. |
Notes
1 HMC provides administration functions to manage a system or group of systems, including the ability to create and alter a partition.
2 A new feature for POWER6, called Shared Dedicated Capacity, enables spare CPU cycles that are "donated" by dedicated processor partitions to be utilized by the shared pool, thus increasing performance and overall system utilization. The dedicated partition maintains first priority for using the dedicated CPU cycles, and sharing occurs only when the dedicated partition has not consumed all of its resources.
3 APV is a paid, separately licensed feature. It is also required for VIOS.
4 APV is a paid, separately licensed feature. It is also required for shared processor partitions.
|
|
