Yeah, that's not good. I suspect the C# HttpServer we're using is not very performant but I'm not aware of any testing
done on that. And as you probably know, there are few alternatives short of directing the CAP directly to a webserver
for processing (which would be a lot of work though and interesting long-term solution for larger installations).
I'm hoping that correcting inventory version numbers will improve matters though that's largely a hypothesis at this
stage. Also, if a cache-cleared viewer is determined to launch large number of requests at once that would still be a
problem.
I wonder if this kind of issue is behind LL's work on improving CAP comms.
On 07/11/12 16:36, Oren Hurvitz wrote:
I've been looking at this issue as well, and it's pretty certain that it's
due to the number of HTTP inventory requests sent from the viewer. I added
logging that shows that when the number of incoming HTTP requests to OpenSim
exceeds about 10/second it starts handling requests very slowly (i.e.,
handling 50 requests at once is slower than handling 50 requests that come
in one at a time). I was actually able to improve performance a bit by
adding delays to make OpenSim accept incoming HTTP connections more slowly!
But that's a dirty hack, and I'm looking for a better solution.
I have found that one reliable source of problems happens when a user
logs-in: the LLLoginResponse packet includes a list of all of the user's
active gestures, and then the viewer sends *separate* FetchInventory
requests for each of those gestures. For example, if the user has 50 active
gestures then the viewer sends 50 HTTP requests at once, and that overwhelms
OpenSim for a while. During that time other requests from the client are
also handled slowly.
I've put in a request to Firestorm (as the leading OpenSim viewer) to change
their behavior so that they'll request all of the gestures at once: see
http://jira.phoenixviewer.com/browse/FIRE-8146 . But they've put this issue
on the "Wishlist", so unless someone steps in (Armin Weatherwax?) this might
never be fixed on the viewer side.
--
View this message in context:
http://opensim-users.2152040.n2.nabble.com/A-bit-confused-tp7578881p7578896.html
Sent from the opensim-users mailing list archive at Nabble.com.
_______________________________________________
Opensim-users mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/opensim-users
--
Justin Clark-Casey (justincc)
OSVW Consulting
http://justincc.org
http://twitter.com/justincc
_______________________________________________
Opensim-users mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/opensim-users
