Virtualizing Resources on the Mainframe
How to virtually get the most from your System z platform.
I/O Resource Virtualization
The System z platform provides two I/O resource-virtualization models used in conjunction with the server's memory and processor-virtualization controls to create LPAR and VM instances. The first model partitions and assigns the server's I/O resources to one or more LPARs or VMs. The second model abstracts I/O resources that don't physically exist, using a different medium, such as system memory, to simulate the I/O resource. The System z platform provides a full complement of I/O resource virtualization controls for both models. The I/O partitioning model is provided by System z firmware and is exploited by both the PR/SM and z/VM hypervisors. The I/O abstraction model is provided by z/VM hypervisor software.
The I/O partitioning model provides dedicated or shared assignment of I/O resources (e.g., I/O channel paths and adapters and their attached I/O resources: devices, LANs, etc.). This partitioning model provides for the creation of multiple logical I/O infrastructures called logical channel subsystems. Each logical channel subsystem is then configured to one or more LPARs to accommodate the partition's I/O processing requirements. The system's I/O resources assigned to each logical channel subsystem may be dedicated to individual partitions or, more typically, dynamically and transparently shared by multiple LPARs. The virtualization controls that accommodate this high-performance I/O resource sharing are embedded within the I/O infrastructure and its attached resources and eliminate the need for hypervisor intervention to accommodate the concurrent use of the I/O resources by the sharing partitions. These controls enable sharing with less than 1-percent virtualization overheads and can significantly improve the overall utilization of the shared I/O infrastructure and its attached I/O resources.
Other server platforms may provide a level of I/O resource sharing by using software-virtualization techniques. However, their capability to provide efficient low virtualization overhead I/O resource sharing is still being developed.
The second virtualization model, deployed by z/VM, is a software model that provides I/O-resource abstractions offering increased I/O-virtualization function and granularity. I/O resources (e.g., System z channel paths, I/O adapters and their attached I/O resources) may be dynamically "manufactured" and deployed to VM instances when no corresponding physical resource is configured. Additionally, they provide virtualization for I/O resources no longer supported to provide compatibility for older OS or software stack requirements. z/VM can create and deploy virtual disks, tapes, consoles, LANs and LAN switches. It can emulate resources, such as the SCSI and Fixed Block Architecture (FBA) disks. It can also partition physical disks into smaller "mini-disks" that provide shared physical disk capacity to individual VM instances.
The System z platform provides I/O resource-virtualization efficiencies not generally available on other platforms, such as the capability to manage and adjust I/O-resource assignment to different LPARs to meet workload demands. For example, the system's Intelligent Resource Director (IRD) provides the automatic reassignment of I/O resources, such as I/O channel paths within LPAR clusters, and provides the capability to reprioritize the execution of I/O operations by each sharing LPAR to accommodate user-specified workload-management goals.
comments powered by