UnixBench WHT variant oddities - anyone else have problems with this benchmark?
I find the unixbench test to give strange results often.
I can bench a freshly installed server sitting on our workbench, in other words sitting idle so nothing running on it to cause odd results; without ahci and get a score. I can then run bonnie++ and record that too. Then I turn on ahci, get nearly double the throughput and even higher IO's with bonnie++, run unixbench again and get the EXACT same score (off by 0.2 maybe).
Single strip of ram vs paired strips interleaved (and thus double the ram too!) gives no performance differences and has no real effect on scoring (again, maybe 0.2 change, sometimes for the worse!).
I'm not extremely familiar with this benchmark but it seems to me that it must be heavily weighted on cpu performance? If I swap out a dual core for a quad core it does make significant difference. But it seems that just about nothing else does, even going from old school 80gb sata to a velociraptor or an intel 32gb ssd. Which produce amazing results when benchmarking mysql/apache etc but seem to make little difference to unixbench...
Our big dual quad core xeons with 16gb ram and 8 x wd velociraptor arrays pull decent scores (1000 ish area) and our low end boxes pull about "roughly" what you'd expect but it's the fact that we can make configuration changes that are quite significant and see little to no positive effect on this benchmark. And as I said, sometimes it even degrades performance according to unixbench. Yet other benchmarks such as bonnie++ show the huge performance improvement.
It's kinda driving me nuts. Anyone else have these problems? Is there some simple work around we are missing? We were hoping to use this benchmark as a simple way for us to internally test and rank our setups.