[clug] Super Dumb question. Network filesystems: NFS, CIFS and ???

[Daniel Pittman]

Adam Baxter <voltagex at voltagex.org> writes:
> On Sat, Apr 24, 2010 at 7:17 PM, steve jenkin <sjenkin at canb.auug.org.au>wrote:
>> Adam Baxter wrote on 24/04/10 6:36 PM:
>>
>> > SSHFS is great until you try to use it on smaller, less powerful devices
>> > like my SheevaPlug - the transfers become CPU bound.
>>
>> In 2000 I got 10Mbps out of a pair of 330Mhz SUN Sparc systems using SSH
>> (1?). They were multi-core machines, wasn't entirely a bust.
>>
>> You can work out the cost-per-bit/bps in watts :-)
>
> Hm, interesting. Given that I'm using a 1.2ghz ARM and a gigabit connection,
> I may be doing something wrong.

Back in the day, two things made a substantial difference to my SSH and
related performance on low-power machines: one was using Blowfish, or another
cypher that was designed to perform well on a 32-bit processor.[1]

The other was to use a cypher that had an optimized assembly language
implementation available for my platform; Linux and the underlying
cryptographic libraries are usually good at exposing those.


I would expect that an ARM system with an optimized implementation could fill
out 10Mbit, at least, using AES.  Finding that implementation could be the
trick, of course...

        Daniel

Footnotes: 
[1]  In my case, x86, but IIRC it targeted ARM-of-the-day as well.

-- 
✣ Daniel Pittman            ✉ daniel at rimspace.net            ☎ +61 401 155 707
               ♽ made with 100 percent post-consumer electrons