CCH, You've identified and interesting problem. You are correct that a single LAS metadata table contains only one product server URL so the problem you are seeing is the one I would expect. We have not experimented with using the localhost URL to get around this. But I would expect you to be able to create a load balanced URL for the product server in the same manner that you created one for the user interface server. Each LAS is truly broken up in to two completely independent servers (UI and Product) that rely on the metadata database to stay in synch. So you should be able to put the product server behind a load balancer the same way you put the user interface server behind one. But there is another solution entirely. Our notion of 'sister servers' is that two servers will divide up the ownership of data, each owning it's own set of associated XML configuration files. All of the requests for dataset A products will be handled by LAS A, dataset B products by LAS B. Setting up LAS 'sisters' means that the LAS installations share the configuration files after the two independent LAS are set up. Then the user interface servers of A and B can present the complete set of data while the product servers associated with A and B only handle requests associated with the data they 'own'. This does not do any dynamic load balancing, but it does scale well and does not run into the problem you have mentioned. I believe this is how the ECCO folks have united the SIO and JPL servers into a single interface: http://www.ecco-group.org:8080/lasxml-org/servlets/dataset In your case I'd try again to get the product server set up as a URL within your load balancer. You can test it independently of the LAS UI by using the ls.pl and lasget.pl scripts in ~las/bin/. If you get that part to work you should definitely be able to set up the configuration files so the UI servers will reference that URL. We would be very interested writing some documentation based on your experience if you get this to work. -- Jon cch wrote:
|