[Tilesathome] changetracker / flohoff-ct prio3 requests

[Matthias Julius]

Florian Lohoff <flo at rfc822.org> writes:

> Hi,
> just for the curious - i am doing some experiments to get the overall
> throughtput of the t at h clients/server up a bit.
>
> The very high numbers of processed tiles e.g. 3-4Ktiles/h are caused by
> a lot of prio4 tiles beeing rendered. Those are mostly empty and the
> clients returns them within 10-20 seconds.
>
> So in case all clients render prio4 requests the server queue overflows
> and is the limiting factor and we see idling clients (stuck in upload
> queue)
>
> In case all clients render prio1/2 requests which are mostly full and
> take time to render the clients are busy but the queue is empty. 
>
> So the idea is to mix those two - so keep the clients busy with prio1/2
> queries and mix prio4 requests e.g. missing tiles into it up to the max
> of the queue.
>
> The culprit i think i have identified is the changedtiles requests which
> seem to run every 6 hours. So we run like 3-4 hours with "pure" prio4 and
> 2 hours with prio2 queries.
>
> The current idea is to queue changedtiles requests all the time e.g.
> every minute but only a few.

I think it would be better if the server would try to keep the active
jobs in a certain ratio.  Maybe 50% prio1, 30% prio2, 15% prio3 and 5%
prio4 or so.  If those 5% prio4 tiles have a quick turnaround this
would get a lot of them processed while most renderers are crunching
on high priority tiles.

Matthias