I'm currently in the process of encrypting my USB 1TB backup drives with LUKS. Part of this process involves filling the disk with random data. For USB2 this takes about 10 hours for a 1TB drive. Admittedly this is a one off process though I often move large chunks of data around for one reason or another. I decided to have a look at USB3. I picked up a Western Digital My Book 3.0 along with a Gigabit PCI-e 1x USB 3 card. I had a little difficulty getting Linux to recognize the drive and found I had to reload the xhci kernel module:
# sudo rmmod xhci_hcd
# sudo modprobe xhci_hcd
Not sure if this issue was a one off or not. I may need to look at this later. Perhaps this module needs to be loaded before/after some of the other USB related modules? Once it was detected I reformatted it with XFS and ran the bonnie++ benchmark on it:
Version 1.03e ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
mercury 12G 88824 99 100197 31 43076 18 55770 94 137263 23 273.3 1
------Sequential Create------ --------Random Create--------
-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
16 7882 67 +++++ +++ 5057 45 7693 69 +++++ +++ 5516 50
Performance was impressive. Looking at block transfers showed writes just below 100MiB/s and reads above 130MiB/s. I didn't test a USB2 drive as I know from experience it would be no where near that speed. I did test my internal storage which is XFS on LVM on RAID10 (4 x 250GB SATA):
Version 1.03e ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
mercury 12G 89567 97 115355 38 58405 32 65973 88 123363 33 536.0 3
------Sequential Create------ --------Random Create--------
-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
16 188 2 +++++ +++ 185 2 189 2 +++++ +++ 173 2
My internal storage performed slightly better than the USB3 drive though it couldn't match the 130MiB/s block read speed. I'm not sure why there was such a difference with the metadata tests. The USB3 device was an order of magnitude faster. It may have been connected with the fact that the USB3 test was a new file system and the internal XFS file system has been in use for quite a while. Not sure though.
Bottom line is that next time I need to transfer my main workstation's file systems to external storage for hardware maintenance it will be a lot faster!