Web lists-archives.com

Re: dd performance test differences




On Fri, Nov 02, 2018 at 01:03:12PM +0000, Adam Weremczuk wrote:
New test results with suggested parameters below:

Slower server
W: 13107200000 bytes (13 GB, 12 GiB) copied, 97.5106 s, 134 MB/s
R: 13107200000 bytes (13 GB, 12 GiB) copied, 28.6353 s, 458 MB/s

Faster server
W: 13107200000 bytes (13 GB) copied, 83.7368 s, 157 MB/s
R: 13107200000 bytes (13 GB) copied, 11.6786 s, 1.1 GB/s

No huge discrepancy but still something making me scratch my head :-)

That's a pretty significant difference--but, there's also a huge difference between the debian versions which means huge differences in drivers and the entire storage subsystem. I'd suggest testing both systems from a current live cd, to take the software out of the equation. If the difference persists, I'd look at the firmware revisions. Also, make absolutely sure everything else is identical (drives, raid config, etc.) before chasing ghosts.

BTW, small values can be useful when e.g. BBU fails and you want to get an idea about performance quickly:

Not really: bs=512 is more a test of the context switching speed of the system than of the disk. At least bs=16k should be used for anything that doesn't involve trying to match block sizes on tapes. You can use a smaller count if you wnat quicker results, but don't use a small block size.