Problem with TCP/IP Port

[Donald Osbourne]

Hi,

My users are experiencing extremely slow response time while using TCP port 5000 (not defined in /etc/services) defined in a 3rd party application, to connect to an Oracle database. The performance is significantly better when an alternate port is assigned.

What is the most common cause of such a problem? And what are the best methods for diagnosing and correcting the problem.

The servers are HP N-Class, running HP/UX 11.0, Oracle 8.1.7 and SunGuard/Camnius Zainet 6.2 application.


Thank you,

Don O.

[Steven E. Protter]

Not all applications are nice and update /etc/services when they use ports. Perhaps there is port contention on port 5000.

I know Oracle doesn't use /etc/services Is there an entry for port 1521 for the database?

SEP

[Sundar_7]

Hi,

I believe, not all the ports are supposed to be defined in /etc/services.

May not be the best suggestion but try installing a sniffer kind and monitor the connections on port 5000.

Thanks,

Sundar.

[Donald Osbourne]

Steven,

Port 1521 is used by oracle, but not defined in the services file.

How can I ping a port to check the response time? And refresh/reset the port is necessary?

[Sundar_7]

There is a quick and dirty way of doing it using telnet

# telnet server port

Services listening on the port may expect different kind of input during the conenction establishment. So this may not work all the time. If u dont get the connection refused message then u r fine.

Resetting the port ? u may have to kill and restart the service listening on the port.

[Michael Steele_2]

'telnet ###.###.###.### 5000'
Connected
telnet>

Where 5000 = port number.

Also refer to 'lsof' with this syntax.

'lsof -i tcp:5000'

See if its installed on your server under /usr/local/sbin or obtain lsof from here:

http://hpux.cict.fr/hppd/hpux/Sysadmin/lsof-4.64/

Search the forum for an already compiled copy to download.

Finally, use netstat and search for dropped UDP packets and collisions.

[doug mielke]

Lanadmin can be used to look for errors, netstat might even give an address of the culprit. but I imagine you've looked there already.

making entries in the services file seems to be only a suggestion, widely ignored. Performance hit on a single port almost has to be contention or compatition for that port. HP cleans up better than other Unix versions after broken connections. In other versions I've had sloppy programmers lock up ports for extended periods, and only cleared after stopping and starting the service being used ( tcp) .

It can also be bogus traffic from almost anywhere that your net allows to the server address, and many network products have tools to identify such. Maybe your network folk can help.

[Kevin Wright]

do you have alot of open connections on port 5000??
netstat -a |grep 5000 will show you.

You may have to decrease your time wait interval if you have alot of ports in this state.

[Brian Bergstrand]

Port 5000 is also used by Windows for some kind of plug-n-play thing (over the network!?). There are known attacks for this service and maybe some script kiddie is trying to attack your machine.

HTH.