[Qgis-psc] (QGIS 2020 Grant Report) QGIS Server and performance monitoring

Andreas Neumann a.neumann at carto.net
Fri Nov 13 08:16:58 PST 2020


Hi,

Ok - I see - the QGIS-Server-PerfSuite, goal is always sequential 
rendering, with no parallel requests - makes sense.

In the Solothurn benchmark we have a docker swarm (or alternatively 
kubernetes) setup with parallel requests. But of course, as you have 
noted, there are other configuration options that are not directly QGIS 
server related. We also wanted to find out what a useful configuration 
for parallel requests is. But there are also overlaps of both benchmarks.

Thanks for the clarification,

Andreas

Am 13.11.20 um 16:12 schrieb Paul Blottiere:
> Hi Andreas,
>
> > Paul: can you clarify if you tested "sequentially" without parallel 
> requests? To me, if I interpret it correctly, this seems to be the 
> case here. Are you doing 20 subsequent, sequential, requests?
>
> Yes there aren't scenarios with parallel requests. As soon as you have 
> parallel requests, you need several QGIS Server instances and a load 
> balancer (because a QGIS Server instance is able to address 1 request 
> at a time). So in this case, you don't test the nominal behavior of 
> QGIS Server, but the response time of the whole "infrastructure".
>
> And in case of the QGIS-Server-PerfSuite, we want to check the 
> rendering time for a specific request.
>
>
> > In the Solothurn benchmarks that Marco did, we almost always did 
> parallel requests - and we found out that the threaded rendering slows 
> down rendering if you have a lot of parallel requests, whereas, if you 
> have no or only 2-3 parallel requests, the threaded rendering is 
> faster. In general, in reality we often have 5-15 parallel requests in 
> our organization in Solothurn.
>
> If you have too many instances of QGIS Server with the parallel 
> rendering activated, then you won't necessarily have enough cores to 
> process the rendering efficiently in case of parallel requests. In 
> this case, it may even lead to performance issues but it's not an 
> issue with QGIS Server itself.
>
>
> Regards.
> -- 
> Paul Blottiere
> QCooperative: https://www.qcooperative.net/
> Hytech Imaging: https://hytech-imaging.fr/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/qgis-psc/attachments/20201113/15876df0/attachment.html>


More information about the Qgis-psc mailing list