TLI t_snd hangs

From: Rybak (rybak_at_caseta.com)
Date: 05/27/04 

Date: 27 May 2004 12:55:13 -0700

Hello all,

I have been searching all over the net for a solution to a problem I
am having with an synchronous t_snd call. The setup of the system(s)
are as follows:

1. Machine A is sending a large amount of data to Machine B
2. Machine A is running lynxOS 3.0.1 and B is running SunOS 5.6

Both machines are using libnsl interface to TLI to communicate over
TCP/IP.

The blocking t_snd occurs about 20 seconds after the network cable has
been removed. Durring those 20 seconds I observe that Machine A's
mbufs are growing very quickly. It is the case that my system waits 25
seconds before it checks the status of the network connection. So what
happens here is the send thread gets blocked on t_snd and the system
is frozen.
  
Question 1. How can I check the real-time status of the transport
endpoint on Machine A. I have tried polling, using t_look and t_sync.
None of those functions have been able to realize that the network has
been disconnected.

>From my searching i have not found a very good method using the TLI
functions to get real-time asynchronous status of the Transport
Endpoint. I have began looking into the kernel for status on either
mbufs or STREAMS availaible.

Question 2. How can I be certain to which kernel resource is being
depleted and causing t_snd to have a "flow control restriction". Can
anyone explain or point me in the direction to find out what is going
on at the kernel level.

Question 3. I am having no luck with using ioctl(conn_fd, I_SETSIG,
S_OUTPUT) to check if stream /dev/tcp is not full. Can someone refer
me to an example on this or shed some light on accessing streams.

Thank you,
Didi

Relevant Pages