Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: A sudden increase of time connexion

Re: A sudden increase of time connexion

From: Billy Verreynne <vslabs_at_onwe.co.za>
Date: Fri, 17 Jan 2003 08:54:37 +0000
Message-ID: <b089b8$gn7$1@ctb-nnrp2.saix.net>


Christophe CAUCHOIS wrote:

> I have an Oracle DB running on a NT server.

What version Oracle? What version NT? What platform? Using MTS or dedicated servers? Network type? (you mention SPX/IPX and TCP/IP)

> Yesterday, the time to connect to this database from a 8.1.7 client was
> about 2 seconds.
> Today, this time connexion can be up to 15 seconds.
> I tested the connexion from various machines and with IP and SPX protocols
> with a TNSPING. Same problem.
> I shut down the server and I found a correct connexion time again.
> But after a few hours, this time increase and come back to about 15
> seconds !

Interesting.

You have not given enough configuration and version details to make any kind of diagnosis as to your problem. But it will not stop me from rambling. :-)

One of the common problems with network connections is when the server (the software configured to accept the network connecs) is configured for DNS or WINS. The server accept the connection. Now it attempts to resolve the IP to a hostname using DNS for example. The DNS server is down. The route long. The DNS server is very busy. Whatever. The resolve takes a few seonds. Longer if it goes unasnwered and times out. In the meantime, the client sits frustrated at his workstation, waiting for the connection to the server to be established.

If you're not using any kind of IP resolving, what happens when you just bounce the listener when things start to slow down?

If this fixes the problem, then there is very likely a problem with the listener or listener config.

If not, then this points to a lower level problem. Possibly in the IP stack or WinNT network configuration.

Other things to do when it slows down is to run traceroutes. How many hops are there? What are the actual times between hops? Are there possibly a slow segment?

Try ping and not just TNSPING. TNSPING makes a connection using a TCP (or SPX) packet. An IP ping sends an ICMP packet. There's a big difference.

If the ping is fast, it means that there are not network bandwidth or speed issues. Which then points a finger at the network service (the listener in this case) as being slow (i.e. the packet transmission speed has been discounted).

I've said it often on this list (and others). Troubleshooting _any_ problem (from a broken toaster to a performance problem on a 80 node RAC cluster), requires you to break down "the system" into smaller components and testing each one of them in isolation. By implication it means that The Big Hammer Approach (tm) can not be used (like bouncing the server to "fix" the problem).

--
Billy
Received on Fri Jan 17 2003 - 02:54:37 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US