transaction rate is pretty high?

Aug 25, 2008 at 1:32 PM

I was struck by the beautiful fade rendering of the tile stream as I roamed in and out on this MultiScaleImage, but began to wonder about how many tiles were being consumed. Out of curiosity I put a counter on the GetTile function to see what kind of render tile counts are streaming in. I did a simple mousewheel zoom from the world to my rooftop and the count was 4536! This means a simple session rolls up 5000 render tiles in no time. I imagine many sessions would be at least twice this amount.

Maybe I am doing this calculation wrong, but if not, the Microsoft license structure is:
"Standard version license is $8000 for 1,000,000/year transactions" at 8 tile renders per transaction this equals only 8,000,000 per year tile renders."

At this rate (5000 tile renders per view session) there would only be 1600 user views per year before kicking up to the $0.01 per transaction cost, which at a 5000 tile render average would work out as:
       $0.01/transaction / 8 rendered tiles/transaction * 5000 tile renders = $6.25 per user view

Is that possible? Am I misinterpreting the GetTile function? Could anybody afford the license for this kind of interface at this rate? It also makes me wonder about the average tile render rate for using the ne ve:Map control. I wish there was a way to count tiles in that control as well.

Aug 25, 2008 at 10:16 PM
A simple way to count tiles in the normal VE control is to find your internet cache, clear it, then navigate around in VE.
Your count is probably correct, and Microsoft needs to address this. Their are some big changes comming up, recall that you're not allowed to hit the tile servers directly.
The good news is that Microsoft is keen to see this techology succeed and although painful I'm sure they will find a way to make it commerically viable.
I wonder if we will see some sort of throttling / reduction parameter for the MSI control at some point, for even typical Gigapixel images if your numbers are correct it will cost a fortune it hosting.
Aug 26, 2008 at 4:00 PM

Hmm I had thought a license would let me hit the tile servers directly but I guess not. I ran into something similar building svg interfaces using Google tiles a couple of years ago.

I am wondering if my GetTile counter is correct. Does every GetTile call result in a tile download from the server? After emptying IE cache, and zooming the DeepEarth interface, I checking the Temporary Internet Files cache where my properties show 399 files. However, my tile counter registered 1588, not even close? I must be setting the tile counter incorrectly or else not every call to "public Uri GetTile(int tileLevel, int tilePositionX, int tilePositionY) " gets into the cache.

Beyond VE or GE tile sources it would be interesting to use a WCS server like Geoserver as the MultiScaleTileSource and see what kind of rendering performance results. I imagine adequate performance would require prebuilding a GeoWebCache tile cache. 

There are plenty of higher resolution (6") aerial imagery sources that are held by local organizations that might like this kind of interface.
Aug 26, 2008 at 11:23 PM
I wonder if it is making many requests for the same tile?
The project is setup for you to easily impliment your own provider, we are fortunate enough that MSFT not stoppping us from using VE for our prototyping even though it is not supported.
If you do impliment another source I'd love to read a blog post or see a sample of it.
Aug 27, 2008 at 4:42 PM
I think you're right John, the DeepEarth GetTile seems to hit VE tiles more than once. I guess, though, that repeat tiles would come from the local cache, which means that the tile cache count is a better predictor of tile access from the VE engine. Of course the VE license precludes direct access to the VE tile engine anyway. Even if this were to change you could be looking at about $1.00 per user view. Assuming an average of 750 tiles per use gives $0.01/8*750= $0.93.

By the way I checked the ve:Map tile use and came up with about 320 for a simple drill down to roof top so it is also going to be costly. A Google Map - GMap2 object gave a little higher tile count at 600, but there is a bit of variation between navigations used.

Besides the Google license is by "page view" ($0.005 per page view) rather than tile render so it appears to compare rather favorably against use of ve:Map. I assume that since the page is unchanged in the course of a GMap2 navigation that it counts as only a single charge against the Google account. If that is true Google looks very good at $0.005 per user view versus VE $0.50 - $1.00 per user view! I hope Microsoft can make some changes to be more competitive.

If I get a chance to work up a test of using DeepEarth against a Geoserver WCS tile engine I'll be sure to write about on my blog.