an adaptive neural network-based method for tile replacement in a web map cache
TRANSCRIPT
-
AN ADAPTIVE NEURAL NETWORK-BASED METHOD FOR TILE REPLACEMENT IN A WEB
MAP CACHE
Santander, Spain, June 20-23, 2011
Ricardo Garca, Juan Pablo de Castro, Mara Jess Verd, Elena Verd, Luisa M. Regueras and Pablo Lpez
Higher Technical School of Telecommunications EngineeringUniversity of Valladolid
-
CONTENTS
Web Map Services
Tiled Map Services
Web Proxy Cache
Neural Network Cache Replacement Policy
-
WEB MAP SERVICES
Server
GetCapabilities
Clients
GetMap
GetMap parameters: Layers Styles Coordinate
Reference System Bbox Width Height Format Transparent BgColor Exceptions Tile Elevation
Map images are generated on the flyflexible, but not scalable
-
TILED MAP SERVICES
Bounding box and scales are constrained to discrete tiles
Community specifications: WMS Tile Caching (WMS-C) Web Map Tile Service (WMTS)
Propietary specifications: Microsoft Bing Maps Nasa World Wind Google Maps
Coarse resolution
Detailedresolution
Level 0
Level l
-
PROXY WEB CACHE
WMS ServerGetCapabilities
Clients
GetMap
DataStoreProxy Cache
-
BRUTE-FORCE APPROACH
Caching the whole map can include millionsof tiles Huge storage requirements Start-time to generate all the content
Many GIS providers lack storage resources
There are map services which update thecartography very often
-
PARTIAL CACHE
When the cache runs out of space it is necessary to determine which tiles should be replaced by the new ones
The cache replacement algorithm proposed in this work uses a neural network to estimate the probability that a request of a tile occurs before a certain period of time.
-
TRAINING DATA
Trace requests from three public nation-widetiled web map services in Spain: Cartociudad, IDEE-Base and PNOA.
1th 7th March 2010 Cartociudad: 25.922 reqs IDEE-Base: 94.520 reqs PNOA: 186.672 reqs
Only cacheable requests considered
-
HEATMAPS
-
REQUESTS DISTRIBUTION
-
NEURAL NETWORK PARAMETERSParameter ValueArchitecture Feed-forward Multilayer Perceptron
Inputs 3 (recency, frequency, size)
Hidden layers 2
Neurons per hidden layer 3
Output 1 (probability of a future request)
Activation functions Log-sigmoid in hidden layers, Hyperbolic tangent sigmoid in output layer
Error function Minimum Square Error (mse)
Training algorithm Backpropagation with momentum
Learning method Supervised learning
Weights update mode Batch mode
Learning rate 0.05
Momentum constant 0.2
-
NEURAL NETWORK INPUTS
-
NEURAL NETWORK TARGET
During the training process, a training record is associated with a boolean target (0 or 1) which indicates whether the same tile is requested again or not in window
Once trained, the neural network output will be a real value in the range [0,1] that must be interpreted as the probability of receiving a successive request of the same tile within the time window.
-
SIMULATION RESULTS
Cartociudad
IDEE-Base
PNOA
-
PROXY CACHE SIMULATION
-
CONCLUSIONS
Serving pre-generated map image tiles from a server-side cache has become a popular way of distributing map imagery on the Web
Storage needs are often prohivitive which forces the use of partial caches
A feed-forward multilayer perceptron can effectively take replacement decisions basedon recency, frequency and size of map requests
-
THANK YOU!
An Adaptive Neural Network-Based Method for Tile Replacement in a Web Map CacheCONTENTSWeb map servicesTILED map servicesProxy web cacheBrute-force approachPartial cacheTraining DataheatmapsRequests distributionNeural network parametersNeural Network InputsNeural Network TargetSIMULATION ResultsProxy Cache SimulationConclusionsThank you!