[kwlug-disc] USB3 expectations? (was: Re: USB3 on Linux)

Chris Irwin chris at chrisirwin.ca
Fri Jun 18 16:40:52 EDT 2010

On Fri, Jun 18, 2010 at 14:31, unsolicited <unsolicited at swiz.ca> wrote:
> So, now is there not only a USB3, there's a SATA3 (6Gbits/s, 600MB/s
> effective)! And most drives can't keep up to SATA 1, begging the question
> ...

As you had mentioned, SSDs are able to saturate SATA-II links already,
so right there is a need (desire) for more speed.

Also note that since SATA-II you can have multiple physical devices
behind a single link (sadly not with all chipsets) whether it be a
simple port multiplier, or a virtual device backed by multiple
physical devices (Drobo, etc).

I was just looking at a self-contained hardware RAID-5 enclosure. It
works with SATA-II drives internally, but exposes itself as a single
(large) SATA-III device. According to what I've been reading, using
eSATA-III actually improve performance, as throughput of n-1 disks is
more than SATA-II can handle (and also apparently more than PCI-E 1x
can too). This is with plain-old 7200 RPM rotational drives. So there
is definitely existing capability for exceeding SATA-II throughput.

> So what does it all mean to you and I, the home consumer, or to business?
> - SCSI will become even less popular, given the higher cost than SATA, as
> SATA performance improves, in (large?) RAID arrays / shelves?

SAS still has benefits for some use cases. Price is not one of those
use cases, though :)


> - even fewer devices than I would wish, like laptops, will come with eSata
> ports in favour of USB3?

Lenovo (and some other manufacturers, apparently) are starting to use
eSATA/USB2.0 hybrid ports. It doesn't take any more room, and laptops
have SATA for the internal hard disk already anyway. I'm not sure if
there are any differences in USB3.0's port design that would cause
problems. I can also see this as a market/price differentiation
feature ("consumers don't need eSATA")


> - OpenWRT devices won't get eSata (typically?) or USB3, the processors not
> (typically?) being able to keep up to the drive? Vis a vis NAS?

Your NAS will throttle on gigabit ethernet before SATA-II.

What do you need that sort of disk throughput on your openWRT device
for (not criticising, just curious)?

> - the benefits of USB3 or SATA3 will largely only be felt by multiple drives
> feeding multiple servers that feed multiple clients (e.g. SQL). Even if the
> clients are all connected Gigabit, the net will still be the limiting
> factor?

I think USB3 will be felt by regular consumers. Flash drives are slow,
usb3-attached SSDs would be awesome. But not at current SSD prices...

SATA-III will also benefit "Small" virtualization users with on-host
storage. "Big" virt users will have centralized network storage (and
thus still benefit according to your rule above), but us smaller guys
running multiple VMs locally will potentially see a benefit.

>  - even on a home connected web server, the limitation will be ISP speed,
> not disk speed? IIRC, typically systems are disk bound, but there's a whole
> lot of 'stuff' between the CPU and disk, preventing an increase in disk
> speed from providing an equivalent increase in system speed? (And this is
> even presuming sufficient requests are coming in sufficiently often to "make
> 'it' worthwhile"?

I use unison to sync $HOME between my laptop and server. It runs on
both local and remote hosts, examines my data on each, then presents a
merge-list (which it executes using an rsync-like method). The part
where my two $HOME directories are being examined takes more time (by
far) than the actual data sync over the network (even using wifi).
Faster disks on my server, even given the 'slow' network link, would
still be a benefit.

> - if drives are SATA, and the bus is USB3, why USB3 instead of just staying
> with (e)SATA?

Getting a USB3 enclosure for your drive would allow backwards
compatibility with USB2 hosts -- granted, most eSATA enclosures also
have USB2.0 ports anyway. Some SATA chipsets don't like hot

eSATA drives need a separate source for power, versus one power+data
cord for USB.

Chris Irwin
<chris at chrisirwin.ca>


More information about the kwlug-disc mailing list