smsbox - too many files error

[fred]

yes I know, the smsbox process reached its file limit.........   but why does it happen?   This is in a simulated environment, kannel on a linux box, message/dlr acceptor web service on a win2k server, and a smsc smpp simulator on a winxp box. Sending 100000 sendsms to kannel. somewhere about 20000 msgs, the smsbox c gets problems sending url request for dlrs. BTW all  100000 MT make it to the smsc simulator   2005-09-21 09:28:08 [11736] [9] DEBUG:   Fragment: (null) 2005-09-21 09:28:08 [11736] [9] ERROR: Error while gw_gethostbyname occurs. 2005-09-21 09:28:08 [11736] [9] ERROR: gethostbyname failed 2005-09-21 09:28:08 [11736] [9] ERROR: System error 24: Too many open files 2005-09-21 09:28:08 [11736] [9] ERROR: error connecting to server `fred3' at port `80' 2005-09-21 09:28:08 [11736] [9] DEBUG: HTTP: Opening connection to `fred3:80' (fd=0). 2005-09-21 09:28:08 [11736] [9] ERROR: Couldn't send request to <http://fred3/traklog/traklog.aspx?remark=sms-serv.......................   my linux file-max is 16380 not sure about per process max files..how to tell?  in any case we shouldn't get to the "Too many files" situation in my opinion,   I reset the IIS service on the win2k in case that was a issue, no, same thing, and still no more messages come after this reset. The LAN is working perfectly. certainly the condition would have started because the web service couldn't keep up, but somehow kannel needs to be able to cope with it. Yes I know it queues up the messages, but somehow this is not enough ? Maybe there is a runaway condition of files (socket requests not closing) - however I think this socket should be getting reused, and connection 'keep-alive' happening, why is it not I ask myself?     I attached to the smsbox process and trying to make sense of how the http requestor works. Is each request a new thread ? maybe not......   I kill kannel, and restart (store file is enabled), 4600 outstanding requests, smsbox immediately goes into the same condition, 'Too many Files' netstat does not show a great number of sockets,so ?? The rate of retries occurs at 1/sec (which is sloow isn't it?)   I would like opinions/hints on how to go about debugging/tracing the problem........ I guess if request fails the socket ought to get closed in send_request().....not conclusive whilst debugging tho.