ENOBUF error on hp-ux 11i

[Anil C. Sedha]

I am facing a issue with ENOBUF all of a sudden. I faced this issue on hp-ux 11.0 and since there is no resolution on 11.0, i had to wait till i upgraded to hp-ux 11i.

The tcp_conn_request_max is set to 4096, which is the only parameter that i have heard of that helps.

We had moved our network connection from a switch to a foundry iron, and now we have reverted back. But the issue which arose after the move to foundry iron, now does not go away even after reverting to the switch.

Can anyone provide their expert suggestions.

Full points will be provided to the resolution, and more important, you'll be helping solve a critical issue.


Thanks
Anil

[James A. Donovan]

How much memory does this system have?
How much free memory is being reported by glance/vmstat?

[Anil C. Sedha]

CPU Util S 57% 50% 57%
Disk Util F 44% 71% 97%
Mem Util S 70% 69% 70%
Swap Util U 42% 42% 42%

besides, i feel this is more of an issue between the OS and the network. As far as i have read, on this, it is not a memory issue. But i am open to all suggestions.

Thanks again

Anil

[Alzhy]

No amount of tuning will completely eradicate this problem. It simply means you've a heavily loaded system - both in CPU and memory terms.

Most often the advice will be to get the applications (server and client) to run faster to avvoid ENOBUF... so it means. add more CPU's if your system is expandable and/or add more memory if your system is somewhat already starved...

[Anil C. Sedha]

Just for troubleshooting.

If i stop all applications except 2 that talk to each other, i should not see this error. Does this sound like a fair statement. A lot of CPU and memory resources will get free then. And i believe whatever the system resources are being used right now, are anyways low enough.

Regards,
Anil

[James A. Donovan]

This link points out that often this error simply means that the client side gave up and closed the connection.

http://unix.derkeiler.com/Newsgroups/comp.sys.hp.hpux/2004-07/0106.html

You may want to try increasing whatever the appropriate timeout values are for your application.

[Anil C. Sedha]

i would agree.

but this issue started all of a sudden. and the developers are not too co-operative. they don't want to make any changes in the code as of yet.

anyone who has faced this issue - can they pls provide some feedback as to what they did for it.

-Anil

[Kent Ostby]

Anil --

First the meaning:

ENOBUFS 233 /* No buffer space available The accept() cannot
complete. The queued socket connect request is
aborted.
*/

PHNE_25134 fixed one case of this for 11.11.

What type of applications are giving this error message ?

[Anil C. Sedha]

Kent,

The application is Cap IA.

Regards,
Anil

[Dietmar Konermann]

Anil,

from my experience the reaon "The queued socket connect request is aborted." is indeed the most likely one. See the comment of HP's Rick Jones in thread http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=237561.

You can trust him. :-)

Best regards...
Dietmar.

[Anil C. Sedha]

Hi Deitmar,

The link is not opening.

Kent - Can you confirm the patch number, i did not find this on hp's patch database.

I agree that the socket is ending before it accepts the socket connection from peer. But my questions are:

1) How can it start happening all of a sudden when there was no change
2) If the socket ends before it accepts the call, i will have to make changes in the application code, and deploy it across all of our systems.

There should be a way out of this.

Anil

[Anil C. Sedha]

Problem is resolved. I had a call in with HP and after it was escalated to a higher level i got the answer (don't know if this is a permanent fix) from Ken (hp engineer).

ndd -set /dev/tcp tcp_early_conn_ind 0


(this makes it turn off). a few of my other systems are working fine with turning it on (1) but i am only concerned with my systems working fine.

Thanks to all who answered. I hope this posting helps those who face this issue in future.

Thanks
Anil