[erlang-questions] [ANN] Erlang/OTP 18.0-rc1 is available for testing.

[Fred Hebert]

On 03/27, Felix Gallo wrote:
>I do indeed like {node(), Integer}!  But it would be highly useful to
>extend monotonic across distribution as well to permit total order and
>comparison; although there are some, ah, implementation details...

Well this is the root of the issue. To have monotonicity over all nodes 
means that you enter the CAP theorem, the idea of 2-phase commits or 
quorum systems.

Those are non-trivial choices, and if, for example, the majority of your 
nodes aren't available, it would be possible for you to never can update 
the monotonic clock again, and may only keep it flat (if not strictly 
monotonic).

That's why you use a node identifier in there -- it lets you compare 
things globally to get a partial prder (I can compare a node's data 
totally, but with other nodes, it's trickier and not always possible).  
This somewhat embraces the idea that time is relative, especially so on 
computers where clocks drift and jump around all the time.