I am working on a Remote Access Server ( pppoe server ) , it is a dedicated box with a Intel Xscale 425 network processor and Montavista Linux as the embedded os. The board has been stabilized for say 512 pppoe sessions, but our ultimate target is for 5k pppoe sessions (long way to go ) but I have started work on the Linux kernel scaling. I scaled the per process descriptors and also the file's open maximum limit , say I have increased the limits for the per process resources. I am not using the pppd , instead we have entire impementation in our application right from the pppoe discovery phase to the ppp session creation. Now after the experimental scaling I am trying to connect nearly 1000 sessions. At some point say after 600 session's have been established the select system call returns with :: bad file descriptor and the print runs on the console indefinitely, the board goes in a stuck state, ( I am also doing a paralell utelnet login to monitor through the shell) the entire thing hangs. What steps can I take to fine tune the kernel for getting the needed scalability, ( a confession here !! I am not knowing my tools right so any suggestion on this ,like what method I should to troublshoot this is welcome too).
Thanks & regards,
Shrikanth R K
