Masó, J., Díaz, P., Pons, X., Monteagudo, J.L., Serra, J., Aulí, F., (2011). Impact of user concurrency in commonly used OGC map server implementations, en: Proceedings of INFOCOMP. Barcelona, October 2011. ISBN: 978-1-61208-161-8.
Impact of User Concurrency in Commonly Used OGC Map Server Implementations
1. INFOCOMP 2011
Impact of User Concurrency in Commonly
Used OGC Map Server Implementations
INFOCOMP 2011
October 2011
Used OGC Map Server Implementations
Joan Masó, Paula Díaz, Xavier Pons, José L. Monteagudo-Pereira,
Joan Serra-Sagristà, Francesc Aulí-Llinàs
Center of Research in Ecology and Forestry Applications
Universitat Autònoma de Barcelona
3. Three Steps for Disaster Management
INFOCOMP 2011
October 2011
GEO-PICTURES is an EC
FP7 SPACE project with the aim of
integrating satellite imagery with in-situ sensors and
geo-tagged images as a tool for decision making in emergency crisis
5. The images
22 satellite images of GeoEye-1 (Orthorectified
GeoTIFF; provided by Google)
(http://www.google.com/relief/haitiearthquake/geoeye.html)
Covering Port-au-Prince
INFOCOMP 2011
October 2011
Covering Port-au-Prince
and surroundings
16-01-2010, 3 days
after the Earthquake
Each image has
196 373 kb 4.21 Gb
40 994x57 392 pixels
pdiaz4
6. Diapositiva 5
pdiaz4 Al Web de descàrrega posa:
By downloading these files, you agree to use the imagery solely for non-commercial use related to emergency relief, and to provide a
proper and distinct photo credit to “GeoEye Satellite Image.”
Això significa que hem de posar el logo de GeoEye a la presentació?
pdiaz; 13/10/2010
7. The service
We are going to test implementations of two standards:
Web Map Service (WMS) standard
INFOCOMP 2011
October 2011
Web Map Tile Service (WMTS) standard
Both standards are Open Geospatial Consortium standards
Assess performance
12. The analysis
Servers ClientsStandardsData
Web Map Service
(WMS)
Web Map Service
INFOCOMP 2011
October 2011
Web Map Service
Cache (WMS-C)
Tile Map Service
(TMS)
Tile Map Tile Service
(WMTS)
13. Traditional WMS server-client interaction
WMS
Server
reques
t GetMap
URL
INFOCOMP 2011
October 2011
respon
se
All studied protocols request maps by creating an URL with specific syntax
http://www.ogc.uab.es/cgi-bin/SIGMA/MiraMon5_0.cgi?VERSION=1.1.0&
REQUEST=GetMap&SRS=EPSG:27573&BBOX=532776,22819,538776,26419&
WIDTH=600&HEIGHT=360&LAYERS=mh-andorra&STYLES=&FORMAT=image/gif&TRANSPARENT=TRUE
URL requests were randomly generated and sent from different clients
The time response is stored in logs and latter analyzed
15. Evaluation of WMS Concurrent Requests to a Single Server
WMS
Server
req
GetMap
INFOCOMP 2011
October 2011
WMS Server Server
res
16. Evaluation of WMS Concurrent Requests to a Single Server
More than one hundred different requests were
done (without optimizing speed configurations).
The influence of the pixel size and the image size in
INFOCOMP 2011
October 2011
The influence of the pixel size and the image size in
the time response were evaluated
The requests were made from up to 5 concurrent
clients.
The time response for the requests are exposed in
graphs.
17. Evaluation of WMS Concurrent Requests to a Single Server
Response time of 5 different server vendors at different scales (pixel sizes) each one under
5 simultaneous requests
10.000
MapServer
GeoServer
MiraMon Server
ArcGIS Server
Express Server
INFOCOMP 2011
October 2011
0.010
0.100
1.000
0.0001 0.0010 0.0100 0.1000 1.0000 10.0000 100.0000
Pixel Size (seconds of arc)
Time(seconds)
18. Evaluation of a Cluster of Servers
To overcome the performance degradation in
concurrent requests a possible solution is to set up a
cluster of servers
INFOCOMP 2011
October 2011
cluster of servers
The cluster of servers act as a virtual single server
6 computers are able to respond at same time to different
clients as if they were like a faster single server
20. Tiling the Request and the Response: Sequential
WMS
Server
req
GetMap
req
GetMap
req
GetMap
req
GetMap
req
GetMap
INFOCOMP 2011
October 2011
Server
21. Tiling the Request and the Response
Some WMS clients are able to tile the space in a regular matrix of small
pieces.
They need several tiles to cover the whole viewport
They can recycle some tiles when the user moves the view laterally
INFOCOMP 2011
October 2011
They can recycle some tiles when the user moves the view laterally
Also can take advantage of the cache mechanisms
If the caching mechanism cannot help the response time can increase even
if each tile is smaller that the whole view
Tiled clients (tiles of 256x256 pixels) were simulated in three
configurations.
22. Tiling the Request and the Response: Sequential
Results of the WMTS speed metrics
Time response for sequential 256x256 tiled requests on a pure WMS server
10
MapServer
GeoServer
Tilecache
INFOCOMP 2011
October 2011
0.01
0.1
1
0.00090.00100.00240.00290.00510.00760.01020.01450.01880.02460.03310.04620.18550.21300.26540.47170.56701.03831.6425
seconds of arc
Seconds(time)
Tilecache
MMServer
ArcGIS Server
Express Server
GeoWebCache
Sequential tiled WMS
23. Tiling the Request and the Response: Concurrent
WMS
Server
req
GetMap 1req
GetMap 2req
GetMap 3req
GetMap 4req
GetMap 5
INFOCOMP 2011
October 2011
Server
24. Tiling the Request and the Response:Concurrent
Results of the WMTS speed metrics
Time response for sequential 256x256 tiled requests on a pure WMS server
10
MapServer
GeoServer
Tilecache
Time response for unlimited concurrent 256x256 tiled requests on a pure WMS
server
10
MapServer
INFOCOMP 2011
October 2011
0.01
0.1
1
0.00090.00100.00240.00290.00510.00760.01020.01450.01880.02460.03310.04620.18550.21300.26540.47170.56701.03831.6425
seconds of arc
Seconds(time)
Tilecache
MMServer
ArcGIS Server
Express Server
GeoWebCache
Sequential tiled WMS
Concurrent Tiled WMS
0.01
0.1
1
10
0.00090.00110.00270.00490.00760.01100.01590.02450.03310.06270.19150.23480.47170.57451.1730
seconds of arc
Seconds(time)
GeoServer
Tilecache
MMServer
ArcGIS Server
Express Server
GeoWebCache
25. Tiling the Request and the Response: Semi-concurrent
Results of the WMTS speed metrics
Time response for sequential 256x256 tiled requests on a pure WMS server
10
MapServer
GeoServer
Tilecache
Time response for unlimited concurrent 256x256 tiled requests on a pure WMS
server
10
MapServer
Time response for up to 4 concurrent 256x256 tiled requests on a pure WMS
server
INFOCOMP 2011
October 2011
0.01
0.1
1
0.00090.00100.00240.00290.00510.00760.01020.01450.01880.02460.03310.04620.18550.21300.26540.47170.56701.03831.6425
seconds of arc
Seconds(time)
Tilecache
MMServer
ArcGIS Server
Express Server
GeoWebCache
Sequential tiled WMS
Concurrent Tiled WMS
0.01
0.1
1
10
0.00090.00110.00270.00490.00760.01100.01590.02450.03310.06270.19150.23480.47170.57451.1730
seconds of arc
Seconds(time)
GeoServer
Tilecache
MMServer
ArcGIS Server
Express Server
GeoWebCache
Semi-concurrent Tiled
WMS
0.01
0.1
1
10
0.00090.00110.00270.00490.00760.01100.01590.02450.03310.06270.19150.23480.47170.57451.1730
seconds of arc
Seconds(time)
MapServer
GeoServer
Tilecache
MMServer
ArcGIS Server
Express Server
GeoWebCache
26. Conclusions (1/2)
The work presented covers:
A metrics on WMS and WMTS services
GeoServer, MapServer, MiraMon Map Server, ArcGIS Server, Express Server
TileCache GeoWebCache
A set of recommendations of Disaster Management
Easy to setup: MapServer
INFOCOMP 2011
October 2011
Easy to setup: MapServer
Easiest configure and update: GeoServer
Fastest: Express Server
The speed tests described are a practical demonstration of the suitability of certain servers
and service configurations in certain domains where reliability of services is imperative
We have seen differences in performance of 2 order of magnitude.
All the analyzed servers have slower performances when the number of simultaneous
clients is increased
A cluster of server dramatically improves performance
p2
27. Diapositiva 25
p2 HE FET DUES OPCIONS PER A LA DIAPO 1/2 DE CONCLUSIONS, UNA MÉS DENSA I L'ALTR AMOOLT MÉS LLEUGERA. TAMBÉ POTS
COMBINAR LA LLEUGERA 1/2 AMB LA 2/2 QUE TROB QUE ÉS DENSA PERÒ QUE ÉS LA QUE VAREM FER A VENÈCIA.
p.diaz; 25/10/2011
28. Conclusions (2/2)
In order to improve performance, some clients request tiles to servers that are not
prepared to serve them
This results on no better performance in some servers
Server optimization for tile requests is needed.
Web clients auto-impose themselves a limit in the number of parallel request
INFOCOMP 2011
October 2011
Web clients auto-impose themselves a limit in the number of parallel request
We saw that this more conservative strategy results on better performance
MapServer and GeoServer with common open source services that do not require
any data preparation process but
their performance is worst than other services that require indexing methods like
MiraMon Map Server
MapServer (based on C++ code) performs better than GeoServer (based on Java
code) under single client requests, but GeoServer is surprisingly faster under
concurrent simultaneous requests.