Impressive, looks like viscurl is a real alternative to local files.
Thanks for the information, I'm certainly going to look further on this.
Jan
On 8/30/2019 6:15 PM, Peter Schmitt wrote:
On Fri, Aug 30, 2019 at 9:48 AM Jan Hartmann <[email protected]
<mailto:[email protected]>> wrote:
Thanks Peter, this is really useful. Do you have any real-world
benchmarks for MapServer that compare regular file access with
vsicurl access, using optimized Geotifs? I've seen the tests for
GDAL at https://trac.osgeo.org/gdal/wiki/CloudOptimizedGeoTIFF,
but what about servering large map-sets over the web?
I don't have any sort of formal benchmarks... but I can provide
anecdotes from my experiences. Random access to a 256x256 block of
extremely large COGs served from fast SSD would have maybe 100ms
response. The same COG accessed with /vsicurl/: First time the tile
is requested about 400ms. GDAL has a least recently used curl cache.
When a tile is cached, subsequent requests can be served maybe about
100ms. My cloud provider is AWS... so access from s3 using MapServer
running in a Docker container on an AWS EC2 instance in the us-east-1
region... and the tile is requested from my local wireless network in
Colorado.
When the data is a COG in s3 and MapServer runs near the cloud
storage, performance is quite good. I've scaled only to dozens of
users. Allegedly s3 can scale to 5,500 requests per second per prefix
in s3.
https://docs.aws.amazon.com/AmazonS3/latest/dev/optimizing-performance.html Presumably
it would scale fairly well by adding more MapServer processes.
_______________________________________________
mapserver-users mailing list
[email protected]
https://lists.osgeo.org/mailman/listinfo/mapserver-users
