| Description | Frequently on the PV mailing lists we get questions about why PV fails to scale up to big problems on a server. These users find that when they have small amounts of data, PV performs fine, but when they scale up PV crashes way before data size should become an issue on their cluster. Surely there are others that give up before querying the mailing lists. Instead, they just report the bad findings to colleagues.
Of course, the problem has nothing to do with PV's scalability. PV scales very well. The real problem is almost always that the server is misconfigured so that it cannot perform parallel rendering. In this case, it always sends all of the data to the client. When the data is big, a crash inevitably occurs.
Crashes are bad. We should be doing everything in our power to prevent this. Rather than crash, PV should live on and give the user the best representation it possibly can. Thus, any time PV is in client/server mode, it should have a threshold for the maximum amount of geometry sent to the client. If that threshold is exceeded and the geometry still needs to be sent, an bounding box is drawn in its place. |