FTP timed out while long transfert from client (MVS) to server (UNIX)

[NGUYEN Sieu]

Hi,

I have a FTP connexion aborted on MVS after a long transfert to Unix. MVS is client. Here you can find my FTP/MVS log on attached text.

I have done some following check :
- PING from MVS to UNIX took less than 15 ms.
- On MVS, "keepalive" is set to 1 mn and "datatimeout" is set to 15 mn.
- Patch level on HP/UNIX for FTP is PHNE_23948.
- A daemon is still present on UNIX after the client on MVS is shut down by timeout. Could this daemon loop ??

Anyone can help ? Thank

[Stefan Farrelly]

Try the ftp again but this time keep a continuous ping command running (preferrably on the MVS end). This has certainly helped me before from timeouts, ping seems to help the ftp processes continue to communicate and thus not suffer from timeouts or network glitches. Stop the ping after the transfer is complete.

[NGUYEN Sieu]

Hi Stefan,

Thank you for reply.

But I'm not sure that is due to bandwidth's problems since all my ping seem take less than 15 ms. I would like to know if there would be any loop occurs on UNIX FTP daemon while these very long transfert.

However, I will try continous ping during FTP and see if it will help.

Cheer !!!

[rick jones]

set/increase/check the logging done by the ftpd on the UX side.

also, try to check the tcp stats on both sides and see if perhaps there was a retransmit timeout. it would be nice to know if the connection which was closed was the control connection (suspect the ftpd idle timeout) or the data connection (suspect a transfer problem).

[NGUYEN Sieu]

Hi Rick,

I'm not sure that 's due to control idle timeout because a "keepalive" signal is send from MVS every 60 seconds and pings took less than 15ms. I suspect instead a data connection problem. It looks like a loop from Unix daemon but I have no informations to confirm. There was neither CPU overload nor disk contention on MVS. Other FTP from the same MVS CPU to other HP/UNIX on the same VLAN worked fine. There are no firewall between ours machines.

I'm very perplex !!!

Thank you for reply

[Victor BERRIDGE]

Hi,
Just thoughts...
Have you looked or tried to adjust your tcp keep alive on the unix box?
like ndd -set /dev/tcp
tcp_keepalive_interval to 780 000

tcp_ip_abort_interval to 2000
?

All the best

Victor

[rick jones]

the idle timeout for the FTP command connection will not be reset by tcp-level keepalive commands. those never get beyond the TCP endpoints.

a command connection idle timeout will be logged to syslog with a high enough logging enabled on the ftpd (see the manpage)

if you suspect the data connection, look for connections being reset in the netstat -p tcp statistics and try to correlate them to the messages you see in the ftp client on the MVS system. after that, you will need to see a packet trace to confirm that the ftp data connection is indeed being aborted. i would suggest tcpdump from www.tcpdump.org - some of the liverpool archive mirrors may have pre-compiled versions.