TIFF performance: overviews and internal tiles: Surprise!

[Rahkonen Jukka]

>Overall, I think this test demonstrates the value of in-memory caching
>of data, but does not really measure performance for large datasets
>of which only a very modest portion will ever be cached when requested.

>Nevertheless, I'm very pleased to see some organized effort to measure
>performance in different scenarios.

>PPS. I think it is also wrong to draw any conclusions about differences of
>1 or 2 percentage points in performance.  Such results should be considered
>as equivelent for all meaningful purposes.

Perhaps we could organize a competition: who can serve the same imagery at fastest rate?  There might be own categories for single image and tile indexes, as well as for uncompressed formats and wavelet compressed (ECW, JPEG2000 and MrSID).  Some easily available, free imagery could be used as raw material.  Landsat GeoCover images for example might be usable. Caching could be avoided by using sliding extents in requests, and result could perhaps be a total time of quite a many individual requests, perhaps one thousand.  Other Open source and commercial image servers could join with pleasure the tough but fair game.

-Jukka Rahkonen