Fluffles wrote: > Ivan Voras wrote: >> Fluffles wrote: >> >> >>> If you use dd on the raw device (meaning no UFS/VFS) there is no >>> read-ahead. This means that the following DD-command will give lower STR >>> read than the second: >>> >>> no read-ahead: >>> dd if=/dev/mirror/data of=/dev/null bs=1m count=1000 >>> read-ahead and multiple I/O queue depth: >>> dd if=/mounted/mirror/volume of=/dev/null bs=1m count=1000 >>> >> I'd agree in theory, but bonnie++ gives WORSE results than raw device: >> > > On what hardware is this? Using any form of geom software RAID?
Xeon 5110, 1.6 GHz, see benchmark results on stable mailing list (It's a pretty fast dual-core CPU, core2-based). It's using geom_mirror and the results are with "split" read algorithm. > The low Per Char results would lead me to believe it's a very slow CPU; > maybe VIA C3 or some old pentium? Modern systems should get 100MB/s+ in > per-char bonnie benchmark, even a Sempron 2600+ 1.6GHz 128KB cache which > costs about $39. Then it might be logical DD gets higher results since Maybe on Linux, definitely not on FreeBSD. I've never seen per-char bonnie++ results on FreeBSD that give more than 1 MB/s. And before anyone starts debugging my setup, I'm not the only one with such results :) I recall there were reports before which show geom_mirror doesn't achieve better performance with split reads.
signature.asc
Description: OpenPGP digital signature
