[kwlug-disc] USB3 expectations?

[unsolicited]

Khalid Baheyeldin wrote, On 06/18/2010 4:40 PM:
> On Fri, Jun 18, 2010 at 2:31 PM, unsolicited <unsolicited at swiz.ca>
> wrote:
> 
>> Khalid Baheyeldin wrote, On 06/18/2010 11:54 AM:
>> 
>>> Something I would be interested in is transfer speed from a
>>> hard drive. Is it really that different on USB3 from USB2, or
>>> will it be the same?
>>> 
>> Help me understand the question.
>> 
>> If USB3 is faster than USB2, why would it be the same? Hw/sw not
>> catching up to the specs yet?
>> 
>> Or are you expecting the performance increase not worth the
>> price?
>> 
> 
> I am not thinking OpenWRT at all here. Just backing up to a USB or
> eSATA drive dock.

OK,  but there should be miles and miles of difference between USB and 
eSata.

> So let us assume that an average disk is around 70 MB/s.
> 
> USB 2.0 is 60 MB/s nominally (same Wikipedia article you reference,
> table further below), and the several year old server that I have
> can't even reach that when backing up. The backup size is 90GB, and
> that should be around 25.6 minutes. However, the real time is
> around 1 hour 20 minutes.

Can you do the same (test) backup, but to an internal or eSata drive? 
i.e. Set a baseline. [For all intents and purposes, there's no 
difference between eSata and Sata - except (sometimes) connectors.)

> So, if USB 3.0 is 300 MB/s (same Wikipedia article and table), and
> the disk is still at 70 or 100 MB/s, then what is the benefit from
> going from USB 2.0 to USB 3.0?

But eSata (2) is, what, already 6 times faster than USB2?

> For other applications, it may be useful. But when moving platters
> are involved, there is no advantage yet.
> 
> I should caution that generalizing from the above may be a mistake,
> because the bottleneck maybe somewhere else. Some observations: the
> backup spikes the CPU considerably, despite this being a dual core
> machine. There is lots of free RAM. It could be wait for I/O that
> is the bottleneck. Could be the USB drivers are inefficient. Could
> be the dump utility's buffering is not good (despite specifying a
> large block size). I am not sure.

Thus the benchmark above.

Is there any encryption involved here, or is this an rsync (therefore 
difference comparisons / calculations going on) type backup?

> There are already USB 3.0 drive docks at the local Canada Computers
> stores.

OK, but if you don't have USB 3.0 on the computer, and in the face of 
eSata, what's the point? i.e. If you have to buy an adapter one way or 
the other.


Just bought a GB USB Nic. Sadly, lost my prior speed test results, but 
will post when I run again. It was something like 9 MB/s 10/100, 3.6 
MB/s 2.4GHz 802.11n 65Mbps, and only ~6.9 MB/s 5GHz 802.11n 300 Mbps. 
i.e. 300Mbps is significantly slower than 10/100 ethernet.

I expect the GB USB Nic, maxing out USB at 480 Mbps to be faster than 
10/100, but we'll see how much USB overhead comes into play. I expect 
it to be faster than 10/100, and slower than Gigabit. Unfortunately 
Gigabit isn't an option on the laptop. [Found out, after the fact, 
just like I found out after the fact that the 802.11n is 2.4GHz only.]

And ... good luck getting more than 65Mbps on 2.4GHz. I saw 130 Mbps 
briefly, once, but with ~30 2.4GHz networks around me, I'll seldom 
ever see it. Off to 5GHz 300Mbps I went - only to be disappointed that 
10/100 was at least 1/3 faster. <sigh>