Today's Limits on Emulation Speed
June 6, 2013
The long-term future of 3000 virtualization looks sunny, in part because there's the remainder of our lifetimes for the Stromasys engine to get faster. Using this month's offerings from the company shows there's plenty of performance to make up, simply to get to HP's 3000 benchmarks from 2003. Migrators won't care like homesteaders will.
When you look at the now-shipping Stromasys Charon product line, it's easy to see the product will run as fast as nearly all of the A-Class and N-Class servers. But HP sold three models of N-Class that are still out of the reach of today's virtualization engine speed. Those models represent the threshold Charon must still break to operate as fast as the fastest, hamstrung HP iron.
Hamstrung is a word from horses and people meaning to cripple. It's an acknowledged practice that the 3000's CPUs were down-clocked by MPE/iX. In some cases, like the lowest-end A-Class, the operating system dialed back the processor by 75 percent. A 440MHz CPU was forced to run at a genuine 110 in the entry-level A-Class. This is one reason why a fully-revamped, 1-processor A-Class HP 3000, with a faster bus, still only ran 70 percent faster than a Series 918. Even at the dawn of a new generation of 3000s, HP was keeping the servers in check.
If a company is considering an emulation scenario on the way to a migration, these limits might not matter. At one Dallas-area e-commerce company, consultant Doug Smith reports the 3000 was moving to archive-system status. A migration was in the wings. But for other companies, hoping to match those three biggest-sized 3000s, June's Charon product line will leave them short of a match.
Migrators might not make up much of the Charon customer base. If they've concluded that a midrange 3000 will do the interim job, however, even companies leaving the platform will have enough horsepower. One of the reasons for this involves adding extra CPUs for the unreleased Charon versions. Unlike HP, Stromasys will support a 6-way or 8-way N-Class.
When it was still a product for sale from the vendor, this kind of top-end N-Class would cost more than $100,000. At that, it was a complex hardware-based server subject to the indignities of iron: CPU boards, specialized disks, proprietary memory -- all might fail in the way physical components do. But you could get very fast, with enough memory in an N-Class 500MHz or 750MHz server.
AICS Research, which sold QueryCalc to hundreds of 3000 sites for more than 25 years, still hosts the comprehensive 3000 Relative Performance Matrix on all MPE servers. The listing of benchmarked systems goes all the way back to the Series 30, as underpowered at the time as the early Series 930 was at the start of the First Generation of PA-RISC CPUs. HP wrapped up its benchmarking with a muddled picture. In 1998 it changed its processor comparison rating to the "HP Performance Unit." Since it was near the time of the e3000 renaming, this became known as the HP EPU.
"In effect," wrote the late, great Wirt Atmar in his notes to his matrix at AICS, "what they did was to begin to use a new set of test suites which they felt were more appropriate to the way that HP 3000s were being used."
The numbers aren't absolute measurements of anything other than the time the various systems take to process one of the several HP test suites, but they do allow you to compare with some accuracy the relative performances that you should expect when upgrading to higher performing system.
You must divide by 10 to get a set of EPUs that make sense in HP's final 3000 product lineup. The Series 918 became a 1.0 when HP recalculated its benchmarks. That means the 4-way 500 and both 750 MHz models at right are coming in at 49.9, 60.6 and 76.8 EPUs, respectively. The Stromasys lineup doesn't go that high today. As Atmar said, those are just numbers from a test suite. Your mileage will certainly vary.
Migration prospects these days are most likely to be smaller customers who didn't see a good business case for leaving the 3000 from 2003 to 2013. They need help from third parties to make their move, and useful advice might be to start the migration while using more reliable iron than HP's 3000 "kit," as the British would say. These people have got what they need from Stromasys today.
However, as the economy rises, some of the larger 3000 homesteaders may be replacing HP hardware that makes their boardroom directors nervous. Emulation can happen even at the same time the big homesteaders make a move -- like Amisys healthcare hosted sites are doing -- to something non-MPE.
The biggest homesteaders don't have a place to go yet to match their top-end performance on today's virtualization engines. However, using extra CPU cores will let Stromasys keep adding CPUs.
Charon-HPA/3000 systems use two cores to emulate each PA-RISC CPU – one core fetches instructions and interprets them on-the-fly into equivalent Intel instruction sequences, while the other core works in parallel, pre- fetching code pages and applying various analysis and optimization algorithms to identify higher-level optimizations.
This More Cores strategy puts some of emulation's limits in the hands of Intel. That's a company that has not been shy about adding cores to its Xeon-x86 line of CPUs, the line that drives Charon. The level of innovation on that line is moving at many times the evolution speed of Intel's other enterprise chip, Itanium.
It's a gamble to guess at future performance needs, so selecting a hardware-based 3000 always involved predicting the lifespan of use -- the headroom for a site's application growth. Virtualization isn't much different in the forecasting, but moving up is a new experience. What makes virtualization attractive is the ease of upgrading to a faster model. HP might have called this a board swap back in the days that it sold hardware. This kind of swap was still on the order of $20,000 for the leap from HP. And that was after you turned back your old processor to HP, so it could be resold. Nobody's going to have to turn back a model of Charon software in order to turn up their performance.