Matthew Toseland wrote:

> The other reason why people don't stay is that it's too much hassle to
> update your node when there has been a mandatory build and you managed
> to miss it. Their node doesn't offer them the option to update, and they
> don't know how to update manually, so they just let it go. The solution
> to this is update-over-mandatory support, or to not have mandatory
> builds, or to include support for downloading a new update from emu in
> the node. I believe we need mandatory builds to debug load balancing,
> if for no other reason. Downloading a new update from emu is possible,
> with sufficient warnings; of course it would put load on the mirrors,
> and of course emu can be spoofed or cracked. Update over mandatory is
> also possible, and nextgens has made some steps towards it in recent
> days (N2NTMs over connections to incompatible nodes). What remains to be
> done is packaging a file into a verifiable bag of keys, allowing nodes
> to tell other nodes about the latest update, transfer the data, and tell
> each other when the revocation key has been cracked. We can mitigate the
> effects of mandatory updates through time-bomb builds (this build will
> become mandatory at time X), but we will always need some.

 From my understanding, the reason this truly *is* an issue is that 
Mandatory builds no longer allow "older" nodes to update off of them, either 
immediately, or after a period of time. Presumably, this is because 
something major enough has changed within the operation of the new build to 
invalidate the prior client's method of connection (though I know this is 
not always the case).

Would a possible solution, then, be to provide a superset "update" protocol 
that was less transient and able to be used even if your node was "out of 
date"? Perhaps this would let a "Fallen behind" node connect to an upstream 
peer to at least pull down the update?

-- 
Ken Snider

Reply via email to