Today's web browsers, their rendering engines and JavaScript interpreters are able to display relatively big amounts of vector data. Moving from DOM rendering (as it was implemented with help of SVG in for examples OpenLayers 2) to Canvas (and further to WebGL -- as we are now having in OpenLayers 3 or Leaflet) enables us to display thousands of complex vector features, with complicated on-client vector data styling. With this possibility, we are facing now new types problems: how to send such amount of data through limited internet connection? If we have closer look at the problem, we can see clearly, that old database paradigm has raised one more time: we can not have all three attributes of data in one pot, but only 2 of them: speed of the delivered data or amount of delivered data or their topicality. If we take this limits into account and decide to deal with big amounts of data in fast way, topicality must be sacrificed. In the talk, we will demonstrate some possible solutions for this problem, using tiled vectors, generalization, aggregation of vector data. Also advantages, disadvantages of various new and popular vector formats, such as GeoJSON, TopoJSON or MapBox will be discussed. Geometric data do not have be rendered all the time in all scales and over whole area of interest, but only necessary portion of them. If displayed in smaller scales, aggregation and generalisation can take place on the server side. That implies, that using vector caching mechanism could be considered as well. But if we need direct interaction of the server input with cached vector data, mechanism for this must be defined as well. Also attribute data have to be transfered separately, if all the optimisation was put in the vector geometries. Also possible steps between cached data and real-time data will be discussed.