Task #1221
closed
Assess the difference in the performance of Thredds with and without nginx
100%
Description
Thredds streams a lot of information to visualisation clients. I'm going to evaluate if nginx is a bottleneck for our use case.
Updated by Gianpaolo Coro over 9 years ago
- Assignee changed from Gianpaolo Coro to Tommaso Piccioli
- % Done changed from 0 to 50
I detected around 60% of performance decrease in visualising maps using thredds with ngix. These results could also depend on the different performance of the portals I used to make the experiments, but the difference is very large and I suppose nginx plays a role in this delay. I used a very high resolution GIS layers to do the experiments.
Furthermore, here is a test you can run to get a geoTiff from this service:
At the present status, if not better explored, I would avoid using ngix in front of Thredds.
Updated by Tommaso Piccioli over 9 years ago
- Assignee changed from Tommaso Piccioli to Gianpaolo Coro
- % Done changed from 50 to 0
I really doubt that the performance differences are due to the proxy ngix.
However, to conduct a proper test we can choose one of these ways, or both:
Temporarily remove the proxy nginx server development
Put both virtual machines on the same server xen with the same storage server
At the moment tomcat port 8180 on the dev server is temporarly open, you can test thredds on dev without going through the nginx proxy
Updated by Gianpaolo Coro over 9 years ago
I have done more precise tests visualising two large layers on port 80 and 8180 alternatively:
http://thredds-d-d4s.d4science.org:8180/thredds/wms/public/netcdf/global-reanalysis-phys-001-004-b-ref-fr-mjm95-gridt__ENVIRONMENT_OCEANS_ELEVATION_1366210702774.nc?service=wms&version=1.3.0&request=GetMap&layers=votemper&styles=&srs=EPSG:4326&CRS=EPSG:4326
http://thredds-d-d4s.d4science.org/thredds/wms/public/netcdf/global-reanalysis-phys-001-004-b-ref-fr-mjm95-gridt__ENVIRONMENT_OCEANS_ELEVATION_1366210702774.nc?service=wms&version=1.3.0&request=GetMap&layers=votemper&styles=&srs=EPSG:4326&CRS=EPSG:4326
The average reponse time is variable because it depends on statistical fluctuations in network performance, thus I made several measurements. The 8180 link results to be 1% faster on average.
If this threshold is neglectable with respect to the requirement of having a proxy in front of the service then we can go forward.
Updated by Gianpaolo Coro over 9 years ago
- Status changed from New to Closed
- % Done changed from 0 to 100
