I've been trying to track down a bug that multiple people have
experienced where clients running multiple bots and doing a lot of
things at the same time will suddenly start using exponential amounts of
memory until all of the system memory is used up under Mono on Linux.
Don't have an answer for that yet but I started by profiling
libsecondlife and found that the login sequence creates a ridiculous
amount of Hashtable objects that hang around in memory. I've rewritten
the login process to use built-in XML parsing in .NET instead of the
XmlRpcCS library which should make memory usage a lot more efficient,
but there is still a lot of data passed back (at least 56KB with a full
request) that doesn't necessarily need to be passed back with every
request. It also explains why the server responds to official clients
faster than libsecondlife ones sometimes. The problem is that some of
this data is necessary for certain parts of the code to work (such as
inventory), but not all clients desire that functionality. My idea was
to add booleans to the Settings class that will automatically request
the needed data at login and also enable or disable that part of the
code, so you could set Client.Settings.USE_INVENTORY = true; which would
request root inventory UUIDs, inventory skeletons, and allow the
InventoryManager to function.
What are your thoughts/ideas on how this could work?
The login rewrite is being tested now and should be in SVN by the end of
the day. It shouldn't break anyone's code but Client.Network.LoginError
has been marked obsolete and replaced by Network.LoginErrorKey (just
contains the error reason such as "god", "key", "presence", or "libsl"
if the error happened within the library) and Network.LoginMessage which
will either be the descriptive error message or the message of the day
depending on whether login succeeds or not.
John Hurliman
_______________________________________________
libsl-dev mailing list
libsl-dev@opensecondlife.org
http://opensecondlife.org/cgi-bin/mailman/listinfo/libsl-dev
