Re: [Monotone-devel] Hash collisions resiliency

[Nathan Myers]

On Wed, Apr 13, 2005 at 08:56:36PM -0700, tekHedd wrote:
> Nathan Myers wrote:
> > On Wed, Apr 13, 2005 at 03:22:14PM -0700, J C Lawrence wrote:
> > So, if you're a billion billion billion times more worried about a 
> > hash collision than about your whole family dying in a car collision, 
> > then maybe monotone isn't for you.
> 
> ... there are two aspects to risk management:
> 
>   2) Understanding what is at risk. Knowing _what_ you lose when you
>      lose), minimizing the loss, and having a plan for dealing with
>      it if it happens.
> 
> ... The unanswered question is: what damage can occur
> before the conflict is detected? I would love to have these questions
> answered:

The answers to your questions are exactly the same as if you happen 
to run monotone on a host with flaky RAM, or if you suffer a perfectly 
normal random bit-flip in a CPU register, or in RAM, or on your bus, 
or on your disk, or in the cable leading to it, or on the TCP/IP
connection you're syncing from.   Each of these is overwhelmingly more 
likely than a hash collision.  Do you have plans for such an event?  
If not, then you'd better prepare for all that stuff, first.  If so, 
then you're also prepared for a hash collision.

Actually, I hope that MT is proof against bad packets delivered anyway.
They happen distressingly often.

Nathan Myers
address@hidden