Christopher Schmidt wrote: > On Thu, Jun 26, 2008 at 10:37:59AM -0500, Stephen Woodbridge wrote: >> An observation. We set up a ka-map style cache that is precached. We >> added 4 servers to the layer definition and it appeared that any given >> tile might get served from any of the servers. IE: there is the >> potential for the tile being cached 4 times, once from each server, >> because the host assignment I think is done on a rotating basis. > > No, it's not. > >> Assuming the observations and assumptions are correct, the optimization >> idea would be to do something simple to ensure that a given tile is >> always requested from the same server so once we have it and its cached >> in the browser we never request it from a different server. > > This is the case currently. > >> One idea for doing this would be to change the rotating host assignment >> to an algorithm like: >> >> host_index = checksum(url_less_host) % num_hosts; >> >> where checksum() is a simple checksum of the url string. > > This is exactly what is done. > >> Would this idea be worth pursuing? > > Were it not already in place: Yes. > > Regards,
OK Cool, I guess I need to go back and have the guys explain why they thought 4 servers was slower to load than 2 servers and have them do some less subjective and more quantitative testing. Based on this I'll assume that they did not clear the browser cache between the tests. <sigh> And I guess I should go read more code before sending email :( Thank you. -Steve _______________________________________________ Dev mailing list [email protected] http://openlayers.org/mailman/listinfo/dev
