A previous IBM Destination z article
outlined the evolution and importance of High Performance Fiber Connectivity (FICON) for System z (zHPF), a vital IBM mainframe technology that’s nonetheless existed largely in the shadows. In October, IBM announced additional zHPF enhancements intended to make it applicable to a larger percentage of client workloads, improving System z scale and resilience. Additionally, the DS8000 enterprise class storage for the mainframe and z/OS increased volume size support to 1 terabyte, a four-fold increase over the prior limit. Finally, DS8000 is also delivering an I/O Priority Manager controlled by the z/OS Work Load manager to deliver a comprehensive goal-driven management solution.
Perhaps most notably, zHPF now supports certain I/O transfers for workloads using QSAM, BPAM and BSAM access methods via new DS8000 I/O commands for exploiting zHPF. Along with the new FICON Express 8s channel, this extended support is intended to provide considerable improved I/O throughput without requiring application changes. Additionally, improved throughput for I/O intensive batch jobs should alleviate batch window pressure.
“Almost everything about this announcement emphasizes significantly improved I/O throughput and reduced channel utilization in different degrees,” says Harry Yudenfriend, an IBM Fellow specializing in mainframe I/O technology. “We’re also extending zHPF up to support additional workloads and heritage access methods to facilitate I/O performance improvements and enhance the value of the DS8000 and z/OS.”
A white paper
highlighting all the performance improvements will be made available online.
List Prefetch Enhancements
List prefetch has been enhanced for zHPF and will provide higher cache hit ratios in the control unit and features faster disorganized index scans with DB2 10—up to 4.9 times faster when using FICON Express 8s and the “Rotate Extents” configuration option. Rotate Extents means a volume allocates its extents, rotating through ranks in an extent pool; it’s both a configuration choice and the default choice, balancing a volume across a set of ranks to avoid hot spots. Combined, DB2 10 for z/OS and zHPF is up to 11 times faster, but with a solid-state disk drive, it’s up to 60 times faster.
“List Prefetch—or Hyper List Prefetch—features new device and control-unit architecture to improve the parallelism and caching algorithms used when reading non-continuous extents on disks,” explains Yudenfriend. “Customers can realize much, much better I/O service times and throughput. New function in the z196 and z114 with the FICON Express 8s channel will further enhance the value of the new list prefetch by reducing the CPU overhead for exploiting the function.”
DB2 v10 makes more extensive use of the new list prefetch method of I/O to realize benefits for DB2 utilities such as RUNSTATS, disorganized index scans and sorted row ID list processing. A detailed description
of the DB2 benefits with the new zHPF list prefetch feature is available.
A New 1 TB Ceiling
DS8000 now supports 1 TB volume sizes and includes support for FlashCopy, Metro Mirror, Global Mirror and zGlobal Mirror replication technologies. The one TB volume size is a huge upgrade designed to promote considerable device number consolidation.
“The previous implementation limit was 221 gigabytes—for extended addressability volumes (EAV) support—so this is a four-fold increase in capacity for z/OS clients in need of additional head room for growth,” says Yudenfriend. “For non-EAV users today, this represents a 20-fold volume increase. In either case, it’s a huge increase without introducing the additional complexity of defining and managing more and more devices.
“All the additional high-performance FICON features that were part of the announcement help complement the increased volume size by scaling the I/O rates and bandwidth to match the growing data density,” he continues. “Basically, we’re scaling everything up, evolving the system to address new data needs that are expanding practically every day, it seems.”
Workload Management Enhancements
z/OS and the DS8000 have featured I/O priority management for more than 10 years, where the workload manager uses System z I/O measurements to identify the work that’s missing goals due to I/O delays and calculated I/O priorities—or I/O velocities. October’s announcement included a comprehensive workload management solution by bringing together DS8000 I/O Priority Manager with Workload Manager (WLM) Host-Based Policy. Important workloads now feature improved disk I/O performance via high-priority work accessing DS8700/DS8800 storage-server resources ahead of lower priority work.
“We’ve built in additional algorithms in the control unit to make it a more comprehensive solution than what the original I/O priority function delivered. So, customers get all the benefits of the new I/O Priority Manager with all the existing Workload Manager infrastructure and host-specified policy goals.”