On Feb 22, 2010, at 18:02, Richard Elling wrote:

On Feb 22, 2010, at 9:46 AM, A. Krijgsman wrote:

Today I received a commecial offer on some external USB 3.0 disk enclosure. Since it was new to me I googled my way to wikipedia and found that the specs say
USB 3.0 should have a 5 Gbit speeds capability.
Could it be an interesting solution to build a very cheap storage area network? ( Ofcourse ZFS in the middle to manage the shares. ) Or is this wishfull (e.g. bad) thinking?

It all depends on if the USB disk honors cache flush commands.

'Cheap' is the keyword - you get what you pay for. I found with some USB drives that ZFS scrubs were showing hundreds of checksum errors, so I ditched the drives straight away. The problem is more the firmware rather than the interface itself; but you won't know until you find it.

On the other hand, I've not found drives with FireWire 800 support to have problems, and in any case, FW800 has a better real-world throughput than USB2. By the time you go much above it, you end up with a single drive's spindle being the bottleneck rather than the bus - though of course, multiple drives will start to fill up a bus anyway.

It's worth noting that USB leeches control from the host computer, so even if the bandwidth is there, the performance might not be for several competing drives on the same bus, regardless of how big the number is printed.

Here's some (old) analysis of ZFS and HFS on USB and firewire:

http://alblue.blogspot.com/2008/04/review-iomega-ultramax-and-hfz-vs-zfs.html

How much that translates over to USB 3, I don't know, but there's a difference between 'theoretical' and 'practical'. It would be good to see what kind of performance numbers you can come up with, or if you run into the same kind of problems with USB that existed for slower models. (Sadly, the planned FW3200 seems to have disappeared into a hole in the ground.)

Alex
_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss

Reply via email to