Hi Stephen
>Let me see if I understand this correct:
>
> Client Router Server
> ------------------------------------
> Linux Linux HPUX Bad
> WinNT Linux HPUX Good
> HPUX Linux HPUX Good
>
>With all Linux boxes version 2.2.something
>
>Is this correct?
Yes, this is right.
>This is slightly different from my situation; I had a Linux client and
>Linux server directly connected to the same switch (3Com Superstack
>II, FWIW)
This is probably a configuration I'm going to try as well: put the cvs
server on a linux box and have it communicate to different clients.
However, the configuration I have now simply suggests that there cannot be
a problem on the IP level as the Linux box as router is working as expected.
>Perhaps if I get time I'll repeat the experiment with (a) the most
>recent Alan pre-2.2 kernel and (b) the most recent 2.4-test kernel...
>
>I'll re-iterate my original request, which was not "it's broke - can
>you fix it" but was "okay, how do I go about tracking this one down?"
At this stage, I am uncertain of the cause of the problem. I am pretty much
sure that the card, drivers and the IP level should work as expected. It is
lilkely that there could be a problem with CVS which gets caught in a kind
of deadlock - however, observing the TCP-packet communication and the
system calls done by both CVS client and server shows always the same
result (I put a printf before and after the read/recv call in the client):
- The clilent is stuck in the read system call on the socket for
reads and waits forever
- The server gets EWOULDBLOCK notifications when writing to the
socket and eventually goes
to sleep
- The last packages sent by the server will be resent roughly
every 30-60 seconds.
The client either responds no packages accepted or the last
packages are accepted by the
client but the system call still blocks.
Regards,
-Thomas
Hello!
172.30.0.3 is HPUX?
Upgrade it, you were so unlucky to purchase some invalid OS release.
Each TCP connection to it from any sane OS will freeze miximum after
656 seconds.
I think this bug has been fixed.
[ Dave: one puzzle less, this bogon wraps his own timestamps at 16 bits. 8) ]
Alexey
>
>Hello!
>
>172.30.0.3 is HPUX?
Hello Alexey,
yes, this is the HPUX.
>Upgrade it, you were so unlucky to purchase some invalid OS release.
>Each TCP connection to it from any sane OS will freeze miximum after
>656 seconds.
Thanks for the hint. This stroke me as well as I left it once the whole
night and the server did not terminate the connection.
>I think this bug has been fixed.
Just out of curiosity, was that bug related to a special configuration with
the HPUX or could it have happened between two Linux boxes as well?
-Thomas
Hello!
> Just out of curiosity, was that bug related to a special configuration with
> the HPUX
This is bug of HPUX, to be more exact.
> or could it have happened between two Linux boxes as well?
No.
Alexey