You are currently on IBM Systems Media’s archival website. Click here to view our new website.

MAINFRAME > Storage > Data Management

System z Innovations Automatically Define Configurations for Greater Availability

The Challenge

To automatically define efficient I/O configurations that helped meet service-level agreements for system performance and availability requires some additional infrastructure.

When new devices are added, clients might not know what data will be placed on them. Also, workloads constantly change, some growing faster than others. Dedicating enough resources to meet all possible requirements simply isn’t affordable. When clients plan I/O configurations they tend to over-configure I/O to avoid constraining mainframe CPU capabilities. So the first step was enabling the system to dynamically tune itself to meet the workload requirements as defined in z/OS Workload Manager (WLM).

To manage I/O priorities, WLM started by providing the infrastructure needed to specify goals, monitor workload against them and dynamically adjust resources and priorities to favor important work that misses goals at the expense of less important work. When z/OS workloads miss goals because of I/O delays caused by resource contention, for example, the OS raises the I/O priorities of more important work. The z/OS I/O Supervisor (IOS) ensures I/O requests are queued and executed in proper order. Channel subsystems manage the work queues in priority order, both initially and when re-driving requests after busy conditions are encountered. System z I/O has been built from the casters up with instrumentation that allows the construction of smart algorithms to manage resources and assign priorities.

FICON* channels also prioritize the execution of I/O requests. This capability was extended to allow z/OS technology to pass WLM-derived I/O priorities to control units through the SAN in a device-independent way. Control units can honor I/O priority by throttling link bandwidth to favor higher-priority I/O requests, optimize access to RAID ranks and prioritize reconnections after resolving cache misses.

Parallel access volumes (PAV) technology was invented to allow multiple simultaneous I/O operations from a single z/OS image to a DASD—while maintaining the capability to measure fine-grained I/O service time components. This reduced the time spent queued in the OS while waiting for device availability. The number of PAV aliases assigned to a logical volume controls the number of simultaneous requests that can be started. WLM could dynamically move PAV aliases among logical volumes to help workloads meet goals when I/O queuing time is the source of delays.

Enhanced virtualization techniques were added to make PAV technology much more responsive to workload demands and to more efficiently utilize System z I/O addressing constructs. This HyperPAV technology can virtually eliminate OS I/O queuing time. HyperPAV assigns PAV aliases to I/O devices as application and middleware I/O requests need them, based on I/O prioritizations assigned by WLM. HyperPAV also provides virtualization of System z I/O addressing so OSs can more effectively utilize the number of alias device addresses available across sharing systems, as every OS image can use the same alias address for a different base device at the same time. Also with HyperPAV, when an I/O request finishes for a device, the next request executed is the highest priority request for the set of devices for that control unit. This provides more comprehensive and effective I/O prioritization and improved efficiency.

The DS8000* I/O Priority Manager extends WLM I/O priority queuing to provide more advanced techniques for managing the response times and throughput through the storage subsystems when running mixed workloads with different service-level requirements.

Dynamic Channel Path Management (DCM) allows I/O configurations to be defined in a coarser fashion. Specified policies indicate how many channel paths can be dynamically added to and removed from control units, to adjust bandwidth as needed. DCM also provides the system with the capability to recognize when failing components expose the system to degraded RAS characteristics and dynamically adjust the configuration to avoid single points of failure or repair.

Harry M. Yudenfriend is an IBM Fellow with Systems and Technology Group, System z and Power who joined IBM in 1980. He was named an IBM Master Inventor in 2001 and has achieved his 33rd invention plateau.



2019 Solutions Edition

A Comprehensive Online Buyer's Guide to Solutions, Services and Education.

Finding the Perfect Fit

IBM System Storage Easy Tier tailors SSDs for your workloads

Encrypt and Protect

IBM Tivoli Key Lifecycle Manager solves security problems and meets new standards

IBM Systems Magazine Subscribe Box Read Now Link Subscribe Now Link iPad App Google Play Store
Mainframe News Sign Up Today! Past News Letters