Re: Software vs Hardware RAID 10?
On Sun, Aug 26, 2007 at 09:00:10AM -0500, Neil Gunton wrote:
> Interesting stuff. So how do you test your pci bus saturation?
I just figure it out by the numbers. Maximum on plain PCI is 33MHz x
32bit = 132MB/s. There is a bit of overhead so expect maybe 100MB/s
actual throughput. 66MHz pci would double that, 64bit PCI would double
it too, and PCI-X can be 100 or 133MHz which gives even more. PCIe has
I believe 250MB/s (in each direction so you can read and write at the
same time unlike PCI) per lane, so x4 would have 1GB/s and x8 would have
2GB/s. Once you get to 2GB/s you are starting to approach the limit of
the cpu bus (athlon 64's hytertransport does 6.4 to 8 GB/s depending on
the chipset and cpu model in each direction). At an effective 1066MHz
by 64bit intel's bus does about 9GB/s (although combined, not each
direction).
> I guess I will try and do a test with bonnie++ using both the hardware
> and software RAID... will that be a reasonable test of the differences?
I often just see what hdparm -t says for each disk, and for the software
raid or hardware raid. Gives a vaque idea of the throughput ability of
the setup, although of course it doesn't test what happens if everything
is in use at once.
--
Len Sorensen
Reply to: