[TriLUG] RAID notion applied to networking

[Pete Soper via TriLUG]

On 5/25/20 1:39 AM, Stephen P. Schaefer via TriLUG wrote:

> On 5/24/20 8:40 PM, Joseph Mack NA3T via TriLUG wrote:
>> On Mon, 25 May 2020, Joseph Mack NA3T via TriLUG wrote:
> Ssh can do layer 2 tunnelling, and I see no reason in principle you
> couldn't put bonding on top of that.  It is conceivable that ssh's
> default compression might make up for the extra processing load, but
> only experimentation would determine that.
>
> https://angrysysadmins.tech/index.php/2019/12/grassyloki/ssh-tap-vpn-using-ssh-to-create-a-layer-2-vpn-between-two-machines/
>
> The only warning is Kernighan's law: "Debugging is twice as hard as
> writing the code in the first place. Therefore, if you write the code as
> cleverly as possible, you are, by definition, not smart enough to debug it."

This really speaks to me, and explains a lot why I was miserable so much 
of the time doing only software development. :-) Sometimes the "clever 
as possible" is replaced by "problem space difficult as all get out".

In an earlier post there was mention of "long ago when dinosaurs roamed 
in 1995". Well, about a generation before that I wrote and debugged an 
HDLC implementation from scratch in TI 990 assembly language with no 
other implementation as a reference. The spec doesn't come with reality 
check remarks, starting with the detail that CRC CCITT could not detect 
all the bit stream errors that occur in the real world with 
standards-complying packets, no matter how hard you wanted it to. Of all 
the code I wrote and debugged in my first career it was datacomm code 
that I found the hardest. You mean a 32 bit CRC might not be adequate 
for a TCP packet? Depends on whether you're using decent hardware or 
using tin cans connected with a piece of string! With the 9600 baud 
RS232 cable way WAY longer than spec that Steve Goldman and I had at 
Ivey's in Charlotte, it was virtual cans and string and the two virtual 
kids on each end were very occasionally hard of hearing. So in addition 
to the HDLC spec my implementation had umpteen flavors of defensiveness, 
with every more detailed sanity checking of the header fields and more 
and more meta-protocol states. Ivey's was deliriously happy in the end, 
but my next move was to stupidly join a datacomm startup. (Steve stayed 
put and became a compiler writer and I rendezvoused with him three years 
later with our roles reversed: he was the lead and I wrote back ends, 
happy to let datacomm fade like a bad dream).

With that in mind, although both the Ubiquiti  edge router post and 
Linux bonding posts are loud Siren songs, I feel like the custom eye guy 
in "Blade Runner" these days, and  "I don''t know such stuff" as 
debugging Ubiquity and bonding configurations as they interact with 
modern Linux, I just do (other stuff). So I'll probably just show my 
wife how to switch between the house router and her phone hot spot.

But this is a fascinating discussion!

Pete

>
>      - Stephen
>