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

Home -> Community -> Usenet -> c.d.o.misc -> Re: Is perfection attainable?

Re: Is perfection attainable?

From: Billy Verreynne <vslabs_at_onwe.co.za>
Date: 31 Jul 2003 23:29:43 -0700
Message-ID: <1a75df45.0307312229.185d556b@posting.google.com>


Phil Singer <psinger1_at_chartermi.invalid> wrote i

> I must say I'm sorry. In my effort to condense 4 months of work I left
> out important information.

So you did Phil.. bad boy you are.. punish you I must <breaking out the leathers, whips, chains and cuffs with en evil smirk on the face>

> The networking group has also claimed that they are innocient.
>
> The vendor for the client hardware (and the switchbox, cable, and NIC
> have already been swapped out "just in case") says there is no problem
> there.
>
> Office politics decree that the fault cannot lie in the client software.
<snipped>

Back with Oracle OPS 7.3 we had the exact same problem. Random connections being lost. However, it was like 2 in every 20 connections at a stage (very common). We brought in network specialists and they did their sniffing bit. We found the reason for the ORA-3113, but not the cause.

Reason - for some obscure reason, a packet send to the client had the FIN bit set. Which meant the connection was now simply just terminated.

The cause? No idea. At a stage the hardware vendor (supplying the cluster for OPS) blamed the clients machines (Win'95 notebooks and desktops). Win'95 networking was not that robust. I mucked about on that side (did several years in network support myself) and could not find anything wrong. Did discover several neat tricks with the Win'95 IP stack though (like how to run multiple stacks on the same ethernet driver).

Then in one month we had a lot of changes. New routers, new segments, cluster upgrade, OPS going to 7.3.4 from 7.3.3 and so on.. and the problem suddenly just disappeared.

I still think it was a bug in the Unix cluster kernel. Only because some of the weirdest damn ones I have ever seen, were due to the Unix kernel. The silly bugs (which do not make you laugh) are usually Microsoft ones. ;-)

> And I am doing it here because (as a long time
> lurker, rare poster) I respect the collective wisdom of this group.

Ah.. but then you should know that it takes more than flattery. Er.. blondes and beers to be precise.

> Sybrand, if this makes you curse, I can read German.

Sybrand is Dutch. So try that instead. I will join in. ;-)

--
Billy
Received on Fri Aug 01 2003 - 01:29:43 CDT

Original text of this message

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