Net services too slow

[Guilherme Belinelo]

Hello,

After using the machine for sometime, the responses to my site accesses start to slow down. It keeps until the page is so slow that the browser timeouts. When it happens, external pings are not replied as well. To "solve" it I need to use /sbin/init.d/net stop and then /sbin/init.d/net start. That way the system starts to work fine again. My machine is a L1000, using HP-UX 64. The patches are up to date and I?m using Oracle 8.0.5, Broadvision Enterprise 4.1 and Netscape Enterprise Server 3.6

[Rhonda Thorne]

Before your system gets so slow that you have to restart net, do you view top? If you notice that netisr is at the top or visible in the 80%+ of top, then you have application or network issues. you can search ITRC knowledge tree for netisr for definition of problems and resolutions.

Hope this is helpful
Rhonda

[Guilherme Belinelo]

Thank you Rhonda, I tried that but I don?t have any running process or comand named netisr.

[Steven Sim Kok Leong]

Hi,

Try checking your network parameters. Most importantly, check your tcp_keepalive_interval via ndd. If your TCP keepalive interval is too long, processes may accumulate and hang inside your server simply because network connections to your server may not time out fast enough when a client abruptly disconnects and your server is unaware that the client has disconnected.

We will however need more collected data and statistics to aid your troubleshooting such as top and ps output.

Hope this helps. Regards.

Steven Sim
Brainbench MVP for Unix Admin
http://www.brainbench.com

[Guilherme Belinelo]

Hi Steven,

Sorry for the delay, I was out of the office.

My tcp_keepalive_interval is 1200000 (20 min) and the tcp_fin_wait_2_timeout is 1200000 (20 min). I changed the nddconf file to keep these values on boot time.

I already have the PHNE_21767 (cumulative ARPA Transport patch) installed.

Attached is the output of netstat -an, ps -ef and top commands.

Let me know if you need more info.

Thanks a lot,

Guilherme Belinelo

[rick jones · Accepted solution]

I have dim recollections of "leaks" in the driver receive queues for some of the interfaces. This would manifest itself as a slowdown of networking, ultimately leading to an unresponsive host. Resetting the interface would clear the problem temporarily.

I would make sure you have the latest driver patch(es) for your system.

The stuff about keepalive intervals was likely not involved, I would suggest you keep your keepalive intervals at their defaults. The FIN_WAIT_2 timer is (IMO) a massive kludge only required when devices inapropriately respond to keepalive probes - TCP already has a mechanism based on "normal" keepalives to cull FIN_WAIT_2 states from completely detached (ie closed) connections. IIRC, the FIN_WAIT_2 timer is an arbitrary timer in the FIN_WAIT_2 state, and the FIN_WAIT_2 state is actually a perfectly valid, receive-only state.

[rick jones]

one more thing - there is no longer a netisr in HP-UX 11. Netisr is only on the 10.20 and earlier releases.