cannot accept any more connections - No buffer space available

[Elif Gius]

Hi all,

sometimes we get the message "cannot accept any more connections - No buffer space available".

The Legato Support suggested to increase following tcp parameter: tcp_conn_request_max up to 16384
The old setting was 4096.

But the error still occure sometimes.
Our system is HPUX 11.11

Netstat statistics shows this:
# netstat -s -p tcp
tcp:
2326437602 packets sent
360691834 data packets (3102444303 bytes)
78110 data packets (104468124 bytes) retransmitted
2053145753 ack-only packets (23983875 delayed)
0 URG only packets
3003 window probe packets
87404503 window update packets
11727252 control packets
405366721 packets received
191191631 acks (for 3108184589 bytes)
188735 duplicate acks
0 acks for unsent data
255959209 packets (2620509696 bytes) received in-sequence
24877 completely duplicate packets (36210528 bytes)
148935 packets with some dup, data (217199439 bytes duped)
14125241 out of order packets (2793774377 bytes)
371812 packets (3615276323 bytes) of data after window
91867 window probes
3370152 window update packets
505 packets received after close
331 segments discarded for bad checksum
0 bad TCP segments dropped due to state change
2258397 connection requests
3589325 connection accepts
5847722 connections established (including accepts)
5877014 connections closed (including 29863 drops)
16144 embryonic connections dropped
185296973 segments updated rtt (of 185296973 attempts)
20987 retransmit timeouts
205 connections dropped by rexmit timeout
3003 persist timeouts
45226 keepalive timeouts
44848 keepalive probes sent
0 connections dropped by keepalive
--> 486 connect requests dropped due to full queue
51382 connect requests dropped due to no listener

for me the 486 drops seems to be not to high; but my backup colleguae still asks me because of this error

I dont't want to increase the parameter tcp_conn_request_max anymore. I don't think this would be the solution.

I think the problem could be, that the client sends a RST to the accept() instead of the final ACK.
So the error messages occures
ENOBUFS: No buffer space is available. The accept() cannot complete. The queued socket connect request is aborted.

Sometimes the backup SW have problems with this. So I think the early connection indication should be turned off.

ndd -set /dev/tcp tcp_early_conn_ind 0

So, am I totaly wrong? Has anyone another idea why the message comes up?

Thanks in advance.

[Muthukumar_5]

Tuning tcp_conn_request_max variable may be a solution to it.

Thread http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=803600 recommeds the same.

But you have to restart the application once you tune tcp_conn_request_max so that effect will come to TRUE.

Try to restart and check with netstat -s -p tcp again.

hth.

[Elif Gius]

I've tuned tcp_conn_request_max from 4096 to 16384.
So the parameter is 4 times higher.
And the message still appears.

[Muthukumar_5 · Accepted solution]

Tunig of that specific parameter is good. But you have to restart the application again for not making the specific problem.

Are you using legato networker?. It will run with the tuned parameter effect when restarting again.

hth.

[harry d brown jr]

Post the output of

kmtune

live free or die
harry d brown jr

[Elif Gius]

NetWorker was restarted ... I'll have a look on the netstat statistics

attached the kmtune output

[Sandman!]

after changing the value of tcp_conn_request_max make sure you'ave saved it in the /etc/rc.config.d/nddconf otherwise it won't survive a reboot.

check with ndd /dev/tcp tcp_conn_request_max to make sure this is the case.

cheers