Figure 7

Performance evaluation for four test cases: the direct sets replicate browsing pattern manually performed by a user over 273 seconds; the randomised requests for each client add a random [0.0; 1.0] value to the requested yaw angle, therefore it forces recomputing the sections and does not use cache. Both direct and randomised have a wait (W) and no wait (NW) version, which either respect the timing of the manual requests (e.g. waits between consecutive requests) or it does requests tile one after the other. a) Average request time per number of concurrent clients, where dots show the individual request times for each client and b) Average running time for a test set on clients.