[lug] Software RAID5 performance

David Anselmi anselmi at anselmi.us
Sat Nov 6 19:04:52 MST 2004

Jeff Schroeder wrote:

> # dd if=/dev/zero of=/tmp/bigfile bs=1024 count=10000
> # time cp /tmp/bigfile /xxx (where xxx is the appropriate drive)
> The results for tmpfs:
> real    0m6.937s
> The results for the boot drive:
> real    0m5.918s
> And the results for the RAID drive:
> real    0m4.537s

I don't think this tests what you want.  All the file blocks are in 
memory (because the file exists in a tmpfs).  That's as good as being in 
cache.  So when you copy them all the kernel does is adjust it's 
accounting to map the original blocks to new (additional) inodes.  You'd 
expect them all to take the same amount of time.

Try mounting each with the sync option and see what happens.  Or better, 
get a disk I/O benchmarker and see what that says.


More information about the LUG mailing list