caching strategy

[schmocker]

Hi, I have tested your docker image using 12k Images (100-200 MB/img). Unfortunately I is not very fast with such hight resolution images. I think your caching strategy is not build to render sub-sub-tiles out of previously requested sub-tiles - so the original image has to be loaded into memory for every request. Is this assumption correct?

I have mounted a azure storage account with the original images and and one for caching to a azure app service with 10 nodes where each has 2 vCPU and 8 GB memory. Still the performance is very bad while zooming into an image with IIIF Leaflet.