[Opensim-dev] [Opensim-commits] r2413 - in trunk/OpenSim/Region: Application Environment Environment/Modules Environment/Scenes Examples/SimpleApp
Adam Frisby
adam at gwala.net
Sun Nov 18 14:54:15 UTC 2007
I implemented a "bool fastMode = " flag in the DB system which should
speed this up a little (will make the one DB call for them all), but
overall we need to replace that protocol. It's badly designed.
Perhaps it's time to commit a OGS V2 comms manager using some of the
ideas I posted earlier?
Adam
Dalien Talbot wrote:
> MW,
>
> For the map, I think we need also to optimize the comms between the
> simulator and grid server. We've done a bit of debugging with nebanon
> (thanks neb! :), what happens is that the client fires off a *lot* of
> MapBlockRequest packets.
>
> Currently for every one of them we fire up an XMLHTTP request to the
> grid. Needless to say that 15+ HTTP requests constitute a small-scale
> DoS (especially since the UGA does not seem to be a super performant
> http server) - this was causing HTTP timeout in some cases (hence
> crash on the client - which i trapped now) - but at the moment
> bringing up the map still causes a major hiccup both for UGA and for
> the sim.
>
> My thought is to time-throttle the requests - on the first
> MapBlockRequest record the requested coordinates, and set up the timer
> callback of, say, 3 seconds. every other MapBlockRequest that comes in
> before the callback fires, does either of the two:
> 1) records the requested rectangle coords in the list (to be used
> during http req)
> 2) updates the "requested rectangle" such that it is a rectangle
> covering all the previous requests.
>
> And resets the timer to 3 sec again.
>
> Once the callback finally fires, the region collates all the data for
> the request, and makes a *single* http request - either with the
> "superset rectangle" (although i guess if the MapBlockRequests can
> have disparate coordinates, this is potentially dangerous) - or -
> ideally - with the list of rectangles (for which the current UGA might
> need to be modded ?) Or, if we want to stay 100% backwards compatible,
> just try to collate as much as we can (say, into up to 5x5 rect
> requests), and the rest just shape not to fire too fast..
>
> While this still does not solve the fundamental problem of scalability
> of the UGA, this should make the things bit better.
>
> I was planning to do that today hopefully, if Nebanon is available for
> his kind help again :-)
>
> /d
>
>
> On Sun, Nov 18 2007 03:11:44 -0800, <mw at opensimulator.org> wrote:
>
>>Author: mw
>>Date: 2007-11-18 03:11:44 -0800 (Sun, 18 Nov 2007)
>>New Revision: 2413
>
>
>>Attempt to get World Map working in Grid mode, will need to be using the grid asset server for it to work correctly and has only been quickly tested in a three region grid.
>
> _______________________________________________
> Opensim-dev mailing list
> Opensim-dev at lists.berlios.de
> https://lists.berlios.de/mailman/listinfo/opensim-dev
>
More information about the Opensim-dev
mailing list