[kwlug-disc] USB 2.0 vs. 3.0 on Linux

[L.D. Paniak]

On 09/22/2014 08:16 PM, Khalid Baheyeldin wrote:
> On Mon, Sep 22, 2014 at 7:59 PM, William Park <opengeometry at yahoo.ca
> <mailto:opengeometry at yahoo.ca>> wrote:
>
>     - What kind of test?
>
>
> Copying around 100GB worth of files, most of them multi-gigabyte in
> size, from one disk to the other. Using rsync, while the files do not
> exist on the target
>
> rsync -av /mnt/disk1/ /mnt/disk2
>  
>
>     - Sequential, random?
>
>
> Sequential.
>  
>
>     - DC or USB powered?
>
>
> USB powered. Both disks are portable (one Toshiba Canvio, the other
> Seagate Expansion.
>  
>
>     - Intel, AMD? -- Intel 3Gbps chipset is better than AMD 6Gbps.  
>
>
> This is an AMD machine.
>  
>
>     You see the difference under heavy load, like 4 VMs in 
>
>     Hyper-V as I'm doing now.
>
>
> Bare metal in my case, no virtualization, and no other load on the
> machine.
>
> Seems there is a partial answer here:
>
> http://www.macworld.com/article/2039427/how-fast-is-usb-3-0-really-.html
>
> Spinning disks can only do 114MB/s (7200 RPM that is). SSDs can do 200
> MB/s.
>
59MB/s sequential read speed is entirely reasonable for a 5400RPM drive
- regardless of how it is connected.

In order to get past the 200MB/s limit with a typical USB3 connection,
one has to be more creative eg.
http://marc-abramowitz.com/archives/2007/02/17/getting-good-performance-out-of-usb-hard-drives-in-linux/
Tuning max_sectors should reduce CPU load (probably your bottleneck for
fast USB transfer - watch "top") and increase sequential performance
efficiency.





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://kwlug.org/pipermail/kwlug-disc_kwlug.org/attachments/20140922/33017027/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: OpenPGP digital signature
URL: <http://kwlug.org/pipermail/kwlug-disc_kwlug.org/attachments/20140922/33017027/attachment.sig>