SATA mirrror performance
João Carlos Mendes Luís
jonny at jonny.eng.br
Sun Sep 16 19:16:37 PDT 2007
I've sent this to -performance, but maybe one of these is the right
forum. Sorry to crosspost. Also, I am not currently subscribed to
these two lists, so please CC me.
I have just installed two 500G SATA discs from Seagate, model
ST3500641AS in an ASUS M2N-E motherboard (nVidia MCP55 chipset). Since
this is a home desktop and I need dual boot, I used nVidia's RAID
technology to create the array, while still using the previous disks for
Well, the suggested device to control these disks, IIRC, is ataraid,
so I went for it. But its performance was incredibly slow. I had less
than half a megabyte per second in a raw transfer (dd). Formating 100G
UFS2 partitions take more than a minute. And all this was drivers
fault, as far as I could notice from "systat -v" output. The array was
operating near 100% capacity. While running newfs, and this I remember
for sure, the array was performing at 4 (four!) transfers per second,
and near 100% load.
I also noticed that ataraid does not integrate with GEOM. Shouldn't
it be by now?
Just to be sure it was no defect in disks, they worked perfectly in
Windows XP. So my solution was to build a whole disk RAID1 device using
gmirror, but now I have two independent and non-interoperational RAID
technologies. Indeed, I am very luck that nVidia's RAID does not use
the same sector as gmirror for metadata, or if they use, that it does
After using gmirror, now I have the RAID in its full performance,
getting over 60Mbytes per second at raw reads, very near the 70MBps from
the specs. Could only be better if we already had NCQ working.
Now the question: Is this expected? Is ataraid somehow deprecated?
If that matters, this has been done on the last week's RELENG_6
source build. The CPU is an AMD Athlon 64 X2 5200+, with 3G RAM.
João Carlos Mendes Luís - Networking Engineer - jonny at jonny.eng.br
More information about the freebsd-geom