5.6.x
I just finished testing the new routines I've been writing for last ~week. Expect release tomorrow.
I also spent time on looking for possible bottlenecks in code. Only real issue left, is amount of simultanous io requests on sockets - system runs too quickly out of buffer space. With minor tweaking of buffer sizes, I was able to get 2500 users online (of which 2300+ were downloading). By limiting sends to few hundred, I'd execpt io to be able to handle 5-10k clients.
ChangeLog:
All
-Change: Another code layer was added to socket handling
-Change: Nearly all blocking (cpu intensive/waiting) actions were moved outside the io threads
-Change: Minor optimizations
-Fix: Having more than one 'context' for directory did not work
Ftp
-Change: io code rewritten
-Change: client job managment redesigned
- Fix: Socket read was posted before socket write on login, which could have caused crash on transfer error
Telnet
-Change: io code rewritten
-Change: client job managment redesigned
-Fix: Socket read was posted before socket write on login, which could have caused crash on transfer error
Http
-Change: Disabled for now
Tcl
-Fix: vfs write, used uid for gid.
|
Hybrid Mode
