Improving I/O Performance with HyperPAV
The new HyperPAV disk technology delivers a one-two punch with improved I/O efficiency and performance and reduced I/O resources required.
Illustration by Campbell Laird
The new HyperPAV disk technology delivers a one-two punch by offering customers improved I/O efficiency and performance while simultaneously reducing the number of I/O resources - or device numbers - they use in IBM* System z* platforms. Unveiled in November 2006, the HyperPAV disk technology is an innovative leap forward for Parallel Access Volumes (PAVs).
HyperPAV's precursor, PAVs, were created as components of IBM storage subsystems and OSs in the late 1990s to improve I/O response times. Designed to allow multiple I/O operations to be processed for a single logical device, PAVs are designed to let customers reduce overall I/O queuing delays to devices that typically are very busy since all I/O operations for a single device are usually processed from a single queue. PAVs are managed by creating multiple addresses for a sole logical device. Figure 1 shows how the heritage PAV design worked.
"PAVs are the method by which IBM achieves I/O parallelism on the System z platform to execute many I/O requests, in parallel, from a single OS image," says Harry Yudenfriend, IBM chief architect for z/OS* I/O, and inventor of the HyperPAV technology. "This is all done while also maintaining our mainframe qualities of service, which includes in-band I/O measurements, workload management, security and data integrity, and so on."
HyperPAV technology takes PAVs a step further. Made possible through the storage capabilities provided by the IBM System Storage* DS8000* storage subsystem for System z processors, mainframe FICON* technology and host OS software, HyperPAV was designed so customers can expect to virtually eliminate the queuing of I/O operations.
The technology behind HyperPAV empowers the I/O supervisor component of z/OS to dynamically assign I/O resources instantaneously, on demand, as the workload requires based on customer-specified goals. This means there's no system-adjustment period required during such peak times as "market-open," after lunch or during batch processing. Figure 2 illustrates how HyperPAV works.
With current support being exclusive to the DS8000 storage series and zSeries* and System z processors (for z/OS 1.6 and above), HyperPAV enhances PAV by enabling applications with the capability to attain equal or superior performance than PAV alone while simultaneously using fewer, overall z/OS resources. In February, IBM also announced z/VM* 5.3 support for HyperPAV.
HyperPAV simplifies performance management and truly achieves an on demand I/O model. It provides a multiplier of resources for every OS image, which get their own view of the alias resources to improve the addressing efficiency of the System z platform. —Harry Yudenfriend, chief architect for z/OS I/O, HyperPAV technology inventor, IBM