PowerVM-enabled virtualization is a means to control costs, improve performance, and enhance IT flexibility.

In a business environment that demands that we all find ways to do more with less, virtualization has been accurately promoted as delivering a number of benefits, including the following:

• Reduce hardware costs.
• Reduce server administration costs.
• Reduce power consumption.
• Reduce cooling requirements.
• Simplify and enhance security.
• More easily facilitate dynamic workload balancing.

Virtualization delivers these benefits by consolidating what were separate physical servers and running them as virtual servers on a single physical machine. The fact that there are fewer physical systems accounts for the reduction in hardware costs and for the lower power and cooling requirements. In addition, centralizing servers on fewer physical systems translates into less machines to manage, maintain, and secure and less geographic and hardware disparity to address. Finally, the net reduction of utility (power and cooling) requirements also reduces the cost of standby power requirements.

The load-balancing benefit comes as a result of being able to readily reallocate processor and other resources among the virtual servers. This is, at best, difficult and typically not possible to do rapidly when running separate physical servers.

One benefit of virtualization that is often not fully recognized is that virtualization makes it easy to rapidly create application environments and deploy new applications in either a test or production environment. Businesses no longer have to order and install new hardware when deploying a new application; now, they can create a new partition on existing hardware and run the application there. This minimizes impact to other applications by allowing new ones to run on their own virtual servers.

In addition, with virtualization, it is not necessary to have separate physical systems for test and development environments (assuming that network I/O performance is not adversely affected). They can be run on their own virtual servers within the same physical machine as is used to run the production virtual severs. The strict isolation of the virtual servers provided by an advanced virtualization engine can practically guarantee that any problems encountered during testing and development will not affect the production virtual servers running on the same hardware.

IBM i and, before it, i5/OS and OS/400 have long provided a platform for server virtualization through their logical partitioning (LPAR) facilities. Each partition can run a separate instance of an operating system. And the partitions are isolated from each other so that if one crashes, the others are unaffected.

What's more, because each partition can run a different operating system, including not just IBM i, but also AIX and Linux, IBM i can virtualize servers for all three operating systems running simultaneously on the same hardware. And each partition can have processor resources dedicated to it, or it can share resources with other partitions.

This support for multiple operating systems is critical for today's businesses because few organizations run a totally homogeneous computing environment. Without multi-operating-system capabilities, the benefits that virtualization delivers would be fragmented within the enterprise.

PowerVM standardizes virtualization functionality within the IBM Power Systems family and enhances this IBM i legacy of virtualization support to provide greater flexibility, more functionality, and higher performance.

Through its micro-partitioning feature, PowerVM supports up to 10 logical partitions per processor core. Thus, for example, an eight-core system can support up to 80 partitions. To optimize load-balancing and power-sharing capabilities, processor resources can be assigned to these partitions at a granularity of 1/100th of a core.

Most of the virtualization work of PowerVM is performed by the PowerVM hypervisor. Type 1 hypervisors, such as the PowerVM hypervisor, are run directly on the system's hardware. This compares with type 2 hypervisors, which are software applications running under an operating system.

The Power of PowerVM Virtualization

A number of PowerVM partitioning features are designed to optimize capacity utilization to improve operational performance while reducing server costs by making better use of existing resources. These features include the following:

• Dynamic Logical Partitioning allows you to move processor, memory, and I/O resources between partitions on the fly to adapt to ongoing fluctuations of the loads experienced in each partition.
• Multiple Shared Processor Pools make it possible to automatically balance processor usage among partitions.
• The Virtual I/O server, a special-purpose partition, can be used to virtualize I/O resources and share them among partitions. This eliminates the need for dedicated network adapters, disk adapters and drives, and tape adapters and drives for each virtual server.
• Shared Dedicated Capacity can be used to reallocate, on an as-available basis, spare CPU cycles from a dedicated partition to a shared pool. However, to ensure that the system continues to meet the defined business requirements, the dedicated partition has an unconditional first-call on those cycles.
• Active Memory Sharing can share memory between partitions as required. Memory is dynamically reallocated on the fly to meet the fluctuating requirements of each partition.
• Live Partition Mobility allows AIX and Linux partitions to be migrated from one physical server to another with little or no disruption to ongoing processes. This feature can be used to minimize downtime when the server must be upgraded or maintained.

Note: Some of the features described above require additional licensing fees and/or hardware costs. In addition, some of the features are not available on certain models. Check the chart showing feature availability on the various Power Systems editions.

Managing Power Systems Virtualization

Two tools can be used to manage the virtualization features of PowerVM: the Integrated Virtualization Manager (IVM) or the Hardware Management Console (HMC).

The IVM is a browser-based interface that allows you to configure virtualization features on a single system. It is included with all entry-level and midrange Power Systems servers. There are no additional licensing fees for the IVM, making it a cost-effective virtualization option for small and medium-sized businesses.

The HMC, which is packaged in an external tower or a rack-mounted server, includes a wider feature set than the IVM. In addition, whereas the IVM can be used to configure and manage virtual servers on only a single physical system, the HMC provides a central point of control for multiple physical Power System servers.

Dual HMCs can manage the same set of servers. Thus, if one HMC goes down, the second one can continue to be used to manage the physical and virtual servers attached to it, thereby minimizing the risk of virtualization monitoring, management, and administration downtime.

The HMC communicates with the Power System servers it manages through a standard Ethernet connection.

You can run either IVM or HMC under IBM Systems Director, which provides a single point of control for a multi-server data center.

PowerVM Advantages

PowerVM offers a number of advantages when compared to third-party, software-based virtualization engines. For one thing, IBM designed PowerVM specifically for IBM Power Systems. This reduces the chance of conflicts between the virtualization engine and the hardware and operating systems. And, as a type 1 hypervisor, PowerVM hypervisor provides tight integration with Power System servers and also allows PowerVM to offer more features and deliver higher performance through fewer layers of software.

Because PowerVM provides firmware-based virtualization, it is better able to isolate individual partitions than is the case with purely software-based, third-party virtualization. The result is that any problems or crashes in one partition are far less likely to affect activity in the other partitions. As a result, you are likely to experience less downtime than would be the case with virtualization software that runs under the operating system.

Furthermore, firmware-based partitioning is better able to avoid performance bottlenecks than software-based partitioning. And the greater isolation of partitions also contributes to enhanced security.

Automated, dynamic resource reallocation features—available with PowerVM, but not necessary with third-party virtualization engines—are critical to optimizing resource utilization. Without the ability to reallocate resources dynamically and, preferably, automatically, the physical server must be sized adequately such that each virtual server running on it can be permanently allocated sufficient resources to handle peak loads.

In contrast, when resources can be reallocated dynamically, the physical server can be sized to provide only adequate capacity for the peak load on the physical server as a whole. Because the timing of peak loads tends to vary among virtual servers, the peak load for the physical server as a whole is less than the sum of the peak loads for the virtual servers that run on it.

For a more detailed comparison of PowerVM and other virtualization options, please see the IBM white paper A Comparison of PowerVM and x86 Virtualization Performance.

Lowering the Incremental Costs of Adding Server Functionality

PowerVM-enabled virtualization makes it easy to lower the total cost (staff time and hardware) to add server functionality. Consider the following:

High Availability

HA solutions work by replicating all critical applications and user and system data from a production server to a hot-standby backup server. When the primary server is unavailable due to planned maintenance or unplanned events, users can be switched to the backup server with little or no downtime.

While the ideal HA solution replicates critical user and application data from a production server to a hot-standby server, in some cases there is value in replicating between partitions on the same hardware, such as to accommodate software migration or upgrades or to maintain a virtual server replica when the backup machine is unavailable. Virtualized hardware also makes HA affordable for some organizations that are not yet able to justify redundant hardware.

Furthermore, in a dual-hardware HA environment, the primary and production servers don't have to be separate physical machines. Instead, they could be virtual servers on the same system. Clearly, this approach will not eliminate the downtime that results from the physical server being unavailable, such as during a hardware upgrade, but it can eliminate considerable downtime caused by other events. For example, because different partitions can run different versions of the operating system, using this configuration, users can be switched to the backup virtual server when it comes time to upgrade the operating system of the primary virtual server. This can almost eliminate downtime during upgrades.

The ultimate HA solution uses two physical servers. This allows operations to continue if the physical server needs to be taken offline, such as when it must be upgraded. If those two servers are geographically separated, this configuration also serves to protect against downtime during a disaster because it is unlikely that both servers will be affected by a single disaster. Then, if the primary server is brought down by a disaster, users can be switched quickly to the remote backup.

With the help of virtualization, a dual-system HA solution, whether those systems are local or remote to each other, doesn't have to be as expensive a solution as it might at first seem.

Backup servers typically use few processor and memory resources when they are running solely as backups. During normal operations, their sole function is normally to accept updates from the primary server in order to keep the primary and backup servers synchronized. The backup server might also be used to run tape backup jobs and read-only batch reporting, but these functions usually aren't as demanding as the production applications and they can normally be scheduled for when the system isn't busy.

In a dual-system HA environment, each physical system doesn't have to serve as only a production server or only a backup server. On each of the two systems, some partitions can contain production virtual servers, while other partitions back up the production virtual servers on the other system, such as is depicted in Figure 1.

Figure 1: Split production and backup loads.

Using this configuration, neither server has to be sized sufficiently to handle the full production load because, during normal operations, neither physical system runs all of the production virtual servers.

When one physical or virtual server goes down or must be taken offline for maintenance, its production load will transfer to the other physical system, but this extra load is usually manageable.

Planned maintenance can be scheduled for when the business is closed (increasingly a rarity in the Web-enabled, globalized world) or when system utilization rates are typically low. Even if one of the physical systems is unavailable, a single physical server may be able to handle the full production load during slow periods. And PowerVM's load-balancing features can automatically reallocate resources to the formerly backup partitions that must assume production roles during these periods.

Unplanned outages can be more problematic because they can happen at any time, including during peak-load times. Thus, when using this balanced production/backup configuration, it is important to identify applications that can be shut down or curtailed in order to provide mission-critical applications with adequate performance when unplanned system outages occur.

Disaster Recovery

There are also scenarios in which one needs to rewind a production database to restore lost files or to create reports coincident with past events, such as to report on production and inventories as of the end of the previous night's manufacturing run. To support these requirements, virtualization reduces infrastructural logistics when setting up a secondary environment in which Continuous Data Protection software can be rapidly deployed to deliver "time-machine-like" functionality with little additional hardware cost or burden on the IT staff.

Data Mining, Archiving, or Generating Reports

Virtualization also allows rapid, cost-effective provisioning of a secondary environment from which CDP-style replication or database transformation tools can create a data warehouse with event and/or transactional integrity that can be used for data mining purposes.

These solutions are not a fantasy that will arrive only in the future. A wide range of existing mature, robust solutions are virtual-machine ready and deliver rapid, successful solutions across data mining and other infrastructural software applications. Eliminating the traditional impediments to creating a new server environment creates new levels of freedom to quickly evaluate options and create a more robust computing environment that works for your business.

Virtually Powerful and Affordable

Virtualization provides the ability to create a highly dynamic computing environment capable of rapidly responding to evolving business requirements. Furthermore, because virtualization is better able to optimize system resource usage, it can lower the total cost of operating a multi-server enterprise data center.

As seen above, PowerVM offers a superior virtualization environment, particularly when compared to virtualization engines that operate as a software application under a host operating system. Any organization that currently runs multiple physical Power Systems servers should at least be evaluating PowerVM-enabled virtualization as a means to control costs, improve performance, and enhance IT flexibility.