Out of the Spotlight
zHPF evolution is an important—yet widely unknown—story
10/3/2011 11:16:45 AM |
By Ryan Rhodes
With most of the buzz surrounding the mainframe’s strong string of consecutive fiscal quarters, coupled with the ongoing release of new zEnterprise systems—other smaller, yet important, mainframe developments have been pushed from the limelight. Such is the case for High Performance Fiber Connectivity (FICON) for System z (zHPF), an enhancement to the z/Architecture, FICON interface architecture and device architectures that has been continually improved upon, albeit without much fanfare, since it was first announced in 2008.
“The vast majority of our mainframe customers are still on z/OS and using some flavor of FICON,” says Harry Yudenfriend, an IBM Fellow specializing in mainframe I/O technology. “Clients believe in the FICON technology and choose to deploy it for Linux even though FCP (fibre channel protocol) is an option. FICON has advantages in terms of end-to-end data integrity checking for all device types transparent to applications and middleware, in-band I/O measurements for capacity planning, accounting and problem determination, autonomic workload management, etc. To support this FICON loyalty, we’ve been kind of on what I call this ‘stealth transition’ or transformation from more traditional FICON protocols. In fact, it’s perhaps been a bit too ‘stealth,’ because now we’re in the position of trying to raise customer awareness about it.”
zHPF is one of several technological milestones that emerged from the ongoing evolution of the mainframe I/O architecture. The evolving I/O story goes back to the copper cables that made up the parallel channel solution (also called parallel original equipment manufacturer information, to the fiber optic cables introduced with the enterprise systems connection (ESCON) solution in 1991. ESCON further evolved into the technically superior FICON, which gave rise to zHPF. It was specifically architected to dramatically improve the mainframe’s online transaction processing (OLTP) workload performance, with I/O service times now being as much as 30 percent faster.
“With our existing I/O paradigm, we’ve maintained 100 percent backward compatibility through FICON,” says Yudenfriend. “Essentially, all applications written since 1964 continue to run using that I/O paradigm. With zHPF, we’ve improved on that paradigm by taking steps to try to streamline things to optimize performance for those applications and middleware that don’t need all the capabilities of the traditional I/O stack—for example program-controlled interrupts and conditional execution. Applications and middleware that use IBM access methods continue to run unchanged.”
zHPF architecture streamlining and I/O performance improvements were achieved by reducing the number of channel command words (CCWs) and information units, which dramatically decreased storage subsystem and FICON channel microprocessor overhead. In general, heritage CCWs were replaced by a transport control word, a change that permits the sending of multiple-channel commands to the control unit as a single entity, rather than a stream of individual commands.
The zHPF enhancements optimize OLTP workloads through a variety of access methods, including DB2, virtual storage access method hierarchical file system, partitioned data set extended and System z file system, 100 percent transparent to applications and middleware. In addition to the I/O performance improvement, zHPF architectural changes provide customers with end-to-end system enhancements that improve overall reliability, availability and serviceability (RAS).
Likewise, FICON scale and efficiency are improved with zHPF. It scales I/O rates and bandwidth to grow with data as its density behind a control unit and device increase. When zHPF first became available, small block transfer I/O rates improved two to three times over previous rates, while increased I/O link bandwidth and I/O device commands also helped boost efficiency. The new FICON Express 8s channel, which became generally available this month, provides four times the improvement for zHPF I/O rates over heritage FICON.
“We’ve significantly enhanced System z resilience in terms of I/O rates,” Yudenfriend says. “If you have significant workloads driving your I/O rates up, with zHPF you can perform up to four times more I/Os per second to handle those kinds of I/O spikes much more easily with the same physical infrastructure and footprint.”
A Pretty Big Deal
According to Yudenfriend, zHPF is a key component of the overall System z I/O roadmap. In addition to the increases in I/O rates and link bandwidth, zHPF complements the z/OS strategies of large volume sizes, scale and workload management.
“We’ve been on a steady transformation to increase the percentage of I/O eligible for conversion to zHPF,” he says. “We’re now on the verge of converting 100 percent of the I/O done by DB2, for example, over to zHPF. So there’s an ongoing story about zHPF that’s actually a pretty big deal, although it’s also all a bit technical. Customers will continue to see zHPF evolve and improve, although it may not receive the kind of coverage of say a new mainframe system or something along those lines.”