On Sep 18, 3:37 pm, Ben <bal..._at_comcast.net> wrote:
> I think I made a post a while back after we upgraded to 10.2.0.2. We
> had horrible performance issues. Sql plans were the same and had the
> same amount of I/O going on, the server just couldn't handle it. Long
> story short the only solution we found after 5 weeks of going back and
> forth with Oracle support was to double the cpu's in our AIX5L quad
> processor machine to an 8 way. This got things back to normal.
>
> Now a couple of weeks later I open up another SR trying to get help in
> finding out why all the sudden we have quite a few ora-3136 errors.
> Never had them before the upgrade now within two days we had about 20
> of them. That was about two months ago now and the errors were limited
> to those two days.
>
> So I found that there were some default settings in the listener.ora
> file that had changed to time out connections if they didn't log in
> within 60 secs. Support tells me to set tracing on those clients that
> were receiving the errors. That's fine, now if I only could tell what
> clients it was.
>
> I ask them how to determine the client and they ask for the alert log
> and the listener log from that day. They come back with two IP
> addresses, which were two of our main application servers. I ask them
> how they came up with those IP addresses and he tells me that they
> were the only two IP addresses in the listener log. I have an
> external table of my listener log and a parsing function that I
> created from something I read on dbazine, so I decided to check that
> because that didn't sound correct. There were 137 distinct IP
> addresses in the log. Now I'm waiting to hear back on why he thinks
> there are only two.
>
> Seems like it's getting harder and harder to get anyone in metalink
> that can actually help.
Unfortunately I don't think you will get many people who don't agree
with you.
Received on Tue Sep 18 2007 - 19:16:35 CDT
