Today's big buzzword when discussing on demand solutions is virtualization. Although LPAR and dynamic LPAR (DLPAR) provide a form of virtualization, it's the introduction of IBM* POWER5* technology that adds true virtualization and provides flexibility in resource allocation while minimizing constraints. Virtualization encompasses two major features: the capability to share physical resources and the capability to move those resources dynamically between workloads or partitions.
POWER5 virtualization is built on the basics of LPAR technology. Unlike POWER4* systems, POWER5 systems are always running under a Hypervisor*, even if all of the resources are assigned to just one partition. The Hypervisor is a piece of firmware that handles time slicing and dispatching for the partition workloads between processors. There's a slight overhead associated with using the Hypervisor because it needs both memory and processor resources to run. Actual overhead depends on the number of workloads and LPARs and the amount of page mapping that occurs. This can be monitored using the lparstat command. By using lparstat -h, it's possible to see the percentage of processor time spent in the Hypervisor, as well as the number of Hypervisor calls.
POWER5 and POWER4 systems running AIX* v5.2 require the Hardware Management Console (HMC) for partitioning. The HMC allows partitions to be created with a minimum granularity of one processor and with I/O devices being assigned at the slot level. However, once the system is running AIX v5.3 or an enabled version of Linux* along with the Advanced POWER* Virtualization (APV) feature, the options change dramatically.
POWER5 technology takes this to the next level with the APV feature. Virtualization in the POWER5 world has several prerequisites. First is the combination of POWER5 hardware and AIX v5.3 or an enabled version of Linux. For the purposes of this article, I'll use AIX v5.3 as a reference. Additionally, virtualization requires the aforementioned APV feature. This comes standard with the pSeries* 590 and 595 servers, but is a fee-based feature on the pSeries 570, 550 and 520 servers. Lastly, the HMC is required to implement virtualization and the new Capacity on Demand (CoD) options.
The key components of virtualization in the POWER5 world include: the POWER5 servers, AIX v5.3 (or an enabled version of Linux), Micro-Partitioning*, simultaneous multi-threading (SMT), Virtual I/O Server, virtual Ethernet, shared Ethernet adapter (SEA) and virtual SCSI. The APV is needed to provide for Micro-Partitioning, SEA, virtual SCSI server and Partition Load Manager (PLM).
Micro-Partitioning changes the whole planning structure for POWER5 servers. Bottlenecks are still identified in the same way and tend to occur in the same places, however, the solutions for these bottlenecks are often different from those in POWER4. The on demand world, implemented in POWER5 with APV, allows greater flexibility and granularity in allocating resources. The ability to add fractions of processors instead of whole processors also helps customers make better use of the servers. The use of Workload Manager (WLM) and PLM helps better optimize those resources.
The key to success in the virtual world is good planning. Examine all of your workloads and determine whether they will play well together and benefit from sharing resources.