Click here for WRQ sponsor message

Database scaling was also outlined at the two-hour briefing from CSY’s Growth Solution team, headed up by Pam Bennett. She said that HP is focusing heavily on I/O and networking, working with 100-Base-T and PCI technologies in its planning. The team’s overall goal is a 20 to 30 percent increase in throughput each year, with a new release available annually.

Bennett also said the Growth team’s work is being executed with an eye toward sustainability – meaning that improvements to the operating system won’t demand that customers change business processes and applications. “We’re making sure that changes or forking of the operating system are internal, to ensure you’re not impacted by the changes we have to make.”

CSY’s Steve Macsisak reported that the overall goal is to get the operating system out of the way of the inherent speed in the PA-RISC architecture. CSY said when it released the first PA-RISC 2.0-based 3000s that tuning of the operating system would harness more of the power in the system’s design.

“We’re trying to get out of the way of the hardware, which can run faster in many cases and reduce bottlenecks and serialization in a lot of cases,” Macsisak said. Enhancements to the 3000’s Transaction Manager (XM), the IMAGE facility that preserves data and prevents corruption, are part of the effort.

Macsisak said such database enhancements are a key part of the Growth team’s efforts. “It doesn’t do us any good to build a lot faster hardware if your applications using TurboIMAGE can’t take benefit of it,” he said.

XM improvements include smoothing the impact when a checkpoint occurs, an improvement that will be available soon to TurboIMAGE customers. CSY is enlarging the user logfile of updates in the XM to lengthen the time between checkpoints. The logfile is divided in half and MPE/iX logs to the file until it fills and then checkpoints, then begins logging to the other half of the logfile.

CSY is also introducing a software branching technique called Flipper that gives MPE/iX a better chance to recover from missed predictions of branches in instruction code logic. These misses cause the system to flush cached code and can lower the overall MIPS ratings of systems.

Engineers continue to work on reducing the number of times the 3000 misses reading from the Translation Lookaside Buffer (TLB), giving the system a better chance of finding code it needs in a ready-storage part of the PA-RISC chip. The 3000 is very sensitive to TLB size, something that CSY has demonstrated as it increased 969 and 939 performance with what’s known as “fat cache” versions of processors using a bigger TLB.

An enhancement targeted for the 6.0 release of MPE/iX is virtual space scan reduction, a way to better track which files on the 3000 contain “dirty pages” that are in memory. Determining which part of a file is in memory is a job executed by the 3000’s memory manager, which locks semaphores while it’s doing the job. The growth team is helping MPE/iX “do a better job of keeping up what data has changed, so we won’t have to search the whole 4-gigabyte address space.”

A new feature in 6.0, which is being backported to a 5.5 release, gives managers a way to assign lower I/O priorities to programs which are doing checkpoint processes that can last 20 to 30 seconds.

“This reduces the impact to your online users who are sitting there waiting for response,” Macsisak said. “They won’t notice that you’ve slowed down the checkpoint, and it has a high impact on the number of people that wait at impedes during a checkpoint. Some people do checkpoints every three minutes, others every three hours, but everybody has checkpoints.”

HP also reported that the latest versions of MPE/iX, PowerPatch 4 or Express 3, provide real potential for performance improvements using the TurboIMAGE prefetch option. Prefetch brings data into memory before locking semaphores, which can take more CPU but allows smoother response time by reducing disk I/O waits.

CSY is also looking for companies to volunteer their systems for measurement, “to help us find the next tuning opportunities and make sure we’re doing the right things when we tune something,” Macsisak said. HP will discuss the measurement results with customers “in an informal way, so maybe you can find out what could be done better [in applications].” Systems will run slower during the collection process, he noted. Contact CSY at steve_ macsisak@hp.com to participate.