[svlug] Totally pooched file permissions

[Daniel Gimpelevich]

On Thu, 01 Feb 2007 11:17:31 -0800, mark weisler wrote:

> On Wednesday 31 January 2007 20:35, Daniel Gimpelevich wrote:
>> On Wed, 31 Jan 2007 17:58:58 -0800, DzM wrote:
>> > So I have multiple possible culprits here. Beta OS on skunky lightweight
>> > device that may (or may not) be able to keep up with the traffic. An old
>> > IDE drive that might be in the process of failing. AutoFS/NFS uncleanly
>> > unmounting things and the server not noticing/fixing/protecting. Could
>> > also just be a power hiccup or something. I _did_ find the slug in a
>> > "off" state yesterday. That seems to imply unclean shutdown leaving the
>> > FS in a horked state (though I thought ext2/ext3 was supposed to take
>> > care of that for most circumstances).
>>
>> Don't forget the Achilles heel of the NSLU2: Hard disks may only be
>> connected via USB.
> Daniel's reply intrigues me.
> 
> With external drives (or other devices for that matter) is USB less reliable 
> than, say, Firewire? (I know a bit about Firewire having a controller...my 
> experience with Firewire is that it is faster but I don't know that I've had 
> reliability problems with it.)
> 
> Is there a distinction between using an external USB-connected hard drive for 
> _data storage_ versus using it to store the operating system?
> 
> Is the NSLU2 doomed to second class status if running with, say, Debian where 
> the OS is on the USB hard disk? (This considers that a native NSLU2 runs from 
> its on-board OS using the external USB drive just for storage.)
> 
> Just curious about potential lessons learned.

I thought I'd field this question to installfest regular Ross Bernheim,
but it turns out he's not subscribed to this list. He gives two very
significant reasons why USB is less suitable than Firewire:
1) The 400Mbps of Firewire is always at least 20% faster than the 480Mbps
of USB. If you're doing anything with the drive other than periodically
saving data and reading it back, this becomes an annoying bottleneck.
2) There are no standards specifying what features USB implementations
must have. In contrast, any Firewire implementation that doesn't have all
the exact same features as any other Firewire implementation isn't
Firewire.

As for lessons learned, you already know that Calvin Wong experimented
with using an NSLU2 running Debian for his personal server. I don't know
whether or not you know that he abandoned that effort in favor of a Cobalt
Qube3 because the exceedingly limited RAM in the NSLU2 forced it to use
USB-based swap way too much, considering the unbearably low throughput of
said USB with respect to that use.