Michael Rogers wrote: > Jano wrote: >> This graph shows the amount of scheduled events (future >> insertions/requests) at each time point, by load. There's always a >> scheduled insertion per publisher, but the ~10 minutes delay to notify >> readers makes these events to pile up as insertion rate increases. > > Damn... > > Would it be more realistic to wait until the insert starts (or maybe > even wait until it completes) before informing the readers? I suppose > that's what real publishers would do...
I'd not say it's, per-se, unrealistic, although there are certainly many use cases to be studied. Should we start categorizing them? Perhaps we could use the wiki.
