Re: Question Regarding MTS Server Behavior

Ronald Rood wrote:
>
> Phil Singer <psinger1_at_chartermi.net> wrote in message news:<3EDB265B.68CCF69E_at_chartermi.net>...
> > We are using a 8.1.7. dedicated server configuration, with
> > ADO managing connection pools. At unpredictable moments,
> > the latency in out TCP network increases to where
> > connections start timing out. When this happens, ADO
> > thinks the sever has dropped it, and never uses that
> > connection again, while Oracle waits for DCD to happen
>
> Best would be to solve the network problem ;-)

Agreed. Due to a variety of business considerations, this cannot be counted on. In other words, only 100% certainty that the network problem has been resolved is acceptable. Therefore, I am looking at solutions which allow us to coexist with a failed network.

> I don't know about the mts behaviour in this situation but why not
> just make the expire_time shorter ? (sqlnet.expire_time=N) N in
> minutes.

Actually, there is a TCP parameter which can do something like this; the effect is to increase the number of retries before the connection is abandoned. We regard this as a last resort, since it will generate more network traffic just when we want less.

The expire_time parameter, IFRC, controls Dead Connection Detection. The problem with letting DCD is that the session gets marked as killed, but is not cleaned up right away, so the potential to run out of sessions still exists. It is another option we are considering.

> It could as well make things a little worse since it is generating
> some extra traffic ...
>
> regards,
> Ronald.

-- 
Phil Singer                |    psinger1ATchartermiDOTnet
Oracle DBA
Remove the Obvious to reply