Board index Unix Programmer

socket hang

socket hang

Post by Yves Delaunoi » Sun, 26 Sep 1999 04:00:00



I'm sending files to a tcp-ip "server" from a tcp-ip "client" program
running on HP-UX.
All's working perfectly, except that when printing big files, the "client"
hangs after about 19 MB (and sometimes resumes after a few ten seconds).
Changing the size of the buffer sent does not solve the problem.
The same client gives no problem on other Unix platforms.
Does somebody have experienced such problem already ? Is there a limitation
on some system resources ?

For now, I don't have any idea where nor why it's*. Any idea's
welcome.


 
 
 
Top

socket hang

Post by Fletcher Glen » Tue, 28 Sep 1999 04:00:00


The symptom is consistent with the occurrence of a lost packet.
Perhaps your target system is too slow to keep up with the
packet flow.  Try to slow down the send and see if the problem
still exists.

--
                Fletcher Glenn


> I'm sending files to a tcp-ip "server" from a tcp-ip "client" program
> running on HP-UX.
> All's working perfectly, except that when printing big files, the "client"
> hangs after about 19 MB (and sometimes resumes after a few ten seconds).
> Changing the size of the buffer sent does not solve the problem.
> The same client gives no problem on other Unix platforms.
> Does somebody have experienced such problem already ? Is there a limitation
> on some system resources ?

> For now, I don't have any idea where nor why it's*. Any idea's
> welcome.




 
 
 
Top

1. epoll, and exceeding maximum sockets hangs process?

I'm having a problem with socket programming on Linux. I have many clients
connecting to a listen()ing socket. If I exceed the number of allowable
sockets for my process, accept() will still return a valid socket number.
The problem is that the entire process then stops receiving socket events.
The second a client disconnects and brings the server's socket count below
its maximum, the events start happening again.

The problem is that just exceeding the socket count hasn't been a reliable
indication for me - that is, I can't just hard-code this limit. First, it
may not be the same on every platform the application will run on. Second,
if a socket (a TCP client) is in any of the termination _WAIT states after
it has terminated, I'll be given the socket back but it still seems to
hang the server process. If I wait for them to expire it goes back to
normal.

What's going on, and is there some way I can identify either:
1) That I've hit the limit, so I can stop calling accept(), or
2) That I've EXCEEDED the limit and am now not getting events?

What I'd REALLY like to do is have the accept() call fail but that's not
happening. Also, why does socket 1027 affect socket 43? Shouldn't socket
43 continue operating normally, even if socket 1027 is past my limit?

Thanks,
Bill

2. cron: change shell

3. close on socket hangs the process.

4. 3002 X/Motif -- pitfalls anyone?

5. write to socket hangs the process

6. 5 problems in Debian: keyboard, alias, halt, pcmcia, sources.list

7. Why the TCP/IP socket hang up in my SS 20 running Solaris 2.4 ?

8. some new groups

9. SCO REL5 socket hang!

10. Avoiding socket hangs.....!

11. Client - Server, Client socket hangs!

12. Would a close on a socket hang a select ?

13. Sockets hang if you set SO_RCVBUF < 1013 ?


All times are UTC
Board index