Faster COBOL, post-migration
November 22, 2005
Duane Percox, one of the founders of K-12 software supplier QSS, has put up numbers that show how much faster COBOL becomes once you move away from the HP 3000 hardware. QSS has been moving its MPE/iX software to the Unix and Linux platforms for several years, patiently testing and selecting the best components to the new non-3000 solution. Percox, who's taught software architecture classes for Interex user group conferences, has a new set of numbers that show the speed you can expect to gain on Intel-based hardware.
I assume everyone is enjoying their migration efforts these days as they make their way off their favorite platform. I recently ran my simple COBOL benchmarks on a Linux production-class server which we just installed here at QSS. (The system is a 2-way Xeon 2.8 Ghz; 6Gb RAM; 2x73-Gb 15k hw/raid-1; 4x36gb 15k hw/raid-5; the Linux distro is SuSE Ent. 9 SP2 for 64-bit) This box is now the fastest system I have tested.
... This system can sort an 800Mb file (10-char key; 80-byte record) faster than an HP 3000 A400-100-110 system can sort a similar 30Mb file. But everyone knew that would be the case...
The QSS updated cobol timings can be found at http://qwebs.qss.com/cobdata.html
The numbers at Percox's test page include some genuine eye-openers, especially if you've been locked into the 3000 hardware world with few comparisons. (To be fair, comparing the 3000's performance to anything has been difficult for many years. During the 1990s HP gave up on benchmarking the system with industry standards.) Luckily, the 3000 community has included curious members like Percox who share their results.
Another conclusion that might be drawn from these test results is that HP-UX is a waystation you might not need on your journey away from the 3000. While the HP rp7410 and rp 2470 HP-UX systems topped all 10 of the 3000s used for comparison in file builds and file sorts, and even outperformed the Intel-based Linux servers — well, the HP-UX servers trailed the Linux systems in the main benchmark: Computing all prime numbers up to 50,000. Sure, HP's HP-UX systems didn't trail the Linux systems by as much as HP's hamstrung N-Class servers. But Linux was six times faster than HP-UX in those tests.
When you chip in the vendor independence that Linux promises, could it be that Linux is the better long-term target to aim your migration toward? We'll look at some technical reports about that tomorrow.