Re: alive e/ken style projects?

> 
> As I pointed out in <https://groups.google.com/d/msg/friam/DxvAXHLakdY/W3MCaq0YMsMJ>,
> leaving timeouts to be implemented at the app level does not work.
I think that the point that you made there is more nuanced than that :-)

For some usecases, it is useful to mask errors. These ucs may (I propose) be characterised loosely:
- emphasis is on monitoring, rather than control based on understanding/interpretation (otherwise the app loses understanding of the current state of what’s being controlled)
- masked error rates and durations for temporary loss are ‘small’ compared to the ucs’ view of the world (otherwise most of the system level behaviour will result from errors, rather than designed behaviour. Effectively, the substrate isn’t up to the job for these ucs)
- the economics of the uc requires solutions that are devliered quickly and are low cost, rather than more robust.

On the flip-side, errors need to be visible, especially  where overall system performance needs to be understood and contolled.  There are many cases of wasted expense throwing hardware at compute problems, where the underlying issue was a poorly configured network.

tc
_______________________________________________
cap-talk mailing list
[email protected]
http://www.eros-os.org/mailman/listinfo/cap-talk