Choosing the faster platform
December 13, 2005
HP reports that choosing a database is one of the most significant decisions a migrating site can make. In the same presentation HP's Kevin Cooper mentioned that Marxmeier Software's Eloquence has racked up very good performance numbers, according to reports HP's received from its migrating customers.
But which platform you deploy Eloquence upon can make a difference. In general, the customer base has praised HP-UX as a faster Eloquence platform than Windows, although that kind of generality makes technical experts cringe. Still, there's this report from a user on the Eloquence mailing list:
Eloquence creator Michael Marxmeier addressed these questions:
In testing our software on both Windows and HP-UX we have found that programs that access an Eloquence database take quite a bit longer on the Windows server. The Windows server we are using has two 2GHz processors and 2GB of memory; our HP-UX server has two 750 MHz processors and 2GB of memory. For example, when running a C program that does a serial read through the database it takes 26 seconds on Windows and 10 seconds on HP-UX. On the HP-UX server we have Enable IPC set to 2, when it is turned off the program takes 22 seconds.
My questions are: 1) The enable IPC option does not seem to be valid on Windows (option not included in config file included w/Windows version; when added to the config file it does not make any difference in processing speed). Is there a way to configure a database server on Windows to get similar benefits that the enable IPC option provides on HP-UX?
2) Is there any info available that shows what specs are needed for a Windows server (CPU, Memory) so that it gets similar performance to a certain HP-UX server when accessing an Eloquence database?
The IPC transport is currently not available on Windows. When we tested this initially, it was not critical for the Windows performance (but made a difference on HP-UX). A quick/simple test with the current version seems to support this.
I’m afraid this [Windows performance specs question] is not easily possible, as both the hardware architecture as well as the OS are rather different. For example, if the CPU architectures are different, the CPU frequency has little significance when comparing CPU performance. For CPU specific performance I would assume the SPEC benchmarks (www.spec.org) might be able to provide a rough overview. However, please keep in mind that IO bandwidth and IO performance might also be rather different.
I did some quick/simple/unscientific test comparing a recent HP-UX box against a Proliant DL360 (dual CPU, SCSI) running Windows. The results seem to indicate that for this test the performance should be roughly equivalent (for a small number of users).
John Pittman of Ryco in Australia chipped in his humble opinion about expecting HP-UX performance from Windows systems:
Sorry, but why would you ever expect a Windows system to be as fast as a UX box at anything? I don’t care how fast the Windows CPU is, or how much RAM it has, its just not going to get there. We run 200+ sessions and up to 40 background jobs on an MPE N4000, on a 1x 220Mhz CPU and 1GB of ram, with about 10 IAMGE databases open, some up to 4GB in size. Windows chassis just don’t have the IO bus speed to match a real computer.