Classic running out of memory... huh? what? <long>

[Jarod Wilson]

On Jun 11, 2009, at 3:23 PM, Thomas Charron wrote:

> On Thu, Jun 11, 2009 at 3:10 PM, Jarod Wilson<jarod at wilsonet.com>  
> wrote:
>> On Jun 11, 2009, at 1:39 PM, Tom Buskey wrote:
>> Typically, no. USB sucks horribly for disk I/O.
>
>  Mostly depends on what your talking about.

Yeah, I left out a "relatively speaking, compared to other external  
disk interfaces", there.

> And the quality of the USB disk/host controller.

Eh. Even the best quality stuff gives relatively meager throughput  
(again, compared with FireWire and eSATA). But its certainly possible  
to go from bad to worse.

>>> USB 2.0 is 480 mbits/s which is probably close to 48 MB/s.
>>> 45 MB/s on gigabit ethernet isn't too bad.
>> (480 Megabit/second) * (1 Megabyte/8 Megabit) = 60 MB/s. But in
>> practice, you'll rarely see much better than about 30MB/s, because  
>> all
>> bus arbitration is done by the host cpu, which is grossly  
>> inefficient.
>> FireWire or (even better) eSATA blows USB out of the water for
>> external disk I/O performance.
>
>  True enough.  :-D  But popping in a USB disk to test it works pretty
> well.

Don't get me wrong, I do think USB disks are often quite useful. Just,  
I'd rather use a better interface if I care about performance. My  
external drives w/USB connections (save USB flash disks) are all USB/ 
FireWire or USB/eSATA (or even USB/FireWire/eSATA), and USB is only  
used when the other routes aren't available.

That reminds me, I need to pick up an eSATA expresscard one of these  
days..


-- 
Jarod Wilson
jarod at wilsonet.com