Gregory's Server@gedvondur You might be interested in the story in this post:
https://www.solipsys.co.uk/new/GracefulDegradation.html?ve13mn
|
5 comments
 @gedvondur The problem in this case was that the comms system couldn't handle the traffic being put in the front end. So somehow either the processing capacity needed to be increased -- which wasn't possible -- or the input needed to be reduced. Slowing it gently across the entire store worked well and was implemented elegantly. @gedvondur The underlying user-facing problem was that flow of data through the system has to be continuous and couldn't be seen to stutter at any point. So making the tills tun a bit slower seems the only thing to do. @ColinTheMathmo @gedvondur @TheGibson that made me think of garbage collection, and the similar trade-offs in play there |
@ColinTheMathmo @thegibson Huh, interesting stuff!
I wonder if a backpressure mechanism would work too. Fibre Channel had a backpressure mechanism that would say "stop, I'm full" when the buffers hit max, because FC was a lossless protocol.