« Experts show how to use shell from MPE | Main | Reducing the Costs in a Major MS Migration »

March 05, 2014

What does a performance index represent?

SpeedometerI want to upgrade our production 959KS/100 system to a 979KS/200, and I see the Hewlett-Packard HP 3000 Relative Performance chart says we'll go from a 4.6 to 14.6. What does that increase represent? For instance, does each whole number (like 4.0 to 5.0) represent a general percentage increase in performance?

Peter Eggers offered this reply.

Those performance numbers are multiples of a popular system way back when, based on an average application mix as determined by HP after monitoring some systems and probably some system logs of loads on customer systems. No information here as to where you are on the many performance bell curves. The idea is to balance your system resources to match your application load, with enough of a margin to get you through to the next hardware upgrade.

N-Class to Series 900 Relative PerformancePeople mention system and application tuning. You have to weigh time spent tuning and expected resource savings against the cost of an upgrade with the system and applications as is.  Sometimes you can gain amazing savings with minor changes and little time spent.  Don't forget to add in time to test, QA, and admin time for change management.

There are a many things to consider: CPU speed and any on chip caching; memory cache(s) size and speed; main memory size and speed; number of I/O channels and bandwidth; online communication topography, bandwidth, and strategy; online vs. batch priorities, and respective time slices; database and file design, access, locking, and cache hit strategies; application efficiency, tightening loops to fit memory caches, and compiler optimizations; and system load leveling.

Since you didn't understand the performance numbers, you might hire a good performance consultant that knows the HP 3000. Of course, look for the "low hanging fruit" fruit first for the biggest bang for the buck, and continue "up the tree" until you lose a net positive return on time invested.

You'll also hear it mentioned that adding memory won't help if the system is IO-bound. That is typically not the case, as more memory means more caching which can help eliminate IOs by retrieving data from cache, sometimes with dramatic improvements. This highlights the need for a good performance guru -- as it is easy to get lost in the details, or not be able to see "the big picture" and how it all fits together.

Aside from Eggers' advice, we take note of the last time HP rated its 3000 line.

At HP World in 2002, HP announced the final new 3000 systems, all based upon the PA-8700 processors. At the high end, HP announced a new N-Class system based upon the 750 MHz PA-8700 processor. The new N4000-400-750 was the first HP e3000 to achieve an MPE/iX Relative Performance Units (MRPU) rating of 100; the Series 918 has an MRPU of 1.

HP contends that the MRPU is the only valid way to measure the relative performance of MPE systems. In particular, they maintain that the MHz rating is not a valid measure of relative performance, though they continue to use virtual MHz numbers for systems with software-crippled processors. For example, there are no 380 MHz or 500 MHz PA-RISC processors. Unfortunately, the MRPU does not allow for the comparison of the HP e3000 with other systems, even the HP 9000.

HP has changed the way it rates systems three times over the life of the HP 3000. During the middle years, the Series 918 was the standard with a rating of 1. In 1998, HP devised a new measurement standard for the systems it was introducing that no longer had the Series 918 at 1. It is under this new system that the N4000-400-750 is rated at 100. Applying a correction factor, AICS Research has rated the N4000-400-750 at 76.8 relative to the Series 918’s rating of 1.

08:36 PM in Hidden Value, Homesteading, News Outta HP | Permalink

Bookmark and Share

Use our search engine to find 20 years
of HP 3000 news and articles



The comments to this entry are closed.