Re: SQL*net and multiple network cards..

Why not try and define two IP gateways on the same subnet as the application servers. (Gateways to the two respective networks that ultimately lead to the server). Then if there was a failure on one of the links, then the route would 'go down' and then dead-gateway detection on the part of the interfaces on the NT. app. servers should mean that the other gateway and network is used instead.

For load-balancing, why not set it all up initially to use both networks as well?

All you need to do is say give 3 of the NT boxes a default gateway of a.b.c.d (Net 1) and the other two a default gateway of x.y.z.a (Net 2).

Then on the first three, give them an additional gateway of x.y.z.a (failover to Net 2) and on the other two give them an additional gateway of a.b.c.d (failover to Net 1).

(Advanced properties of the TCP/IP Config screens in NT allow you to do this).

Then whichever link goes down, all NT boxes will get their IP packets routed on to a gateway and route that is still up.

As to the Tnsnames.ora file, well it needs to have the server's IP address and only that. So I don't see how you could change that. The server's IP address doesn't change whether or not one particular link to it is up or not. Not sure if you can specify an explicit route for your IP packets using any particular networking software (except perhaps traceroute / tracert which can be told to go through a particualr router using --somthing-- based routing (can't remember).

If you do manage to create two IP gateways to which the NT boxes can be pointed, each gateway the first and only way to connect to the respective network, then you can basically let the redundancy and fail-over capabilities of good-old I.P. and I.C.M.P. to do all the work for you.

David P. OCP (DBA) MCP (TCP/IP)

Glasgow, Scotland.

You could

rjstevens_at_dera.gov.uk wrote:
>
> Hello
>
> I have recently taken on part of the management of a system in which 5
> NT application servers (running Peoplesoft FS/HR with oracle 8 client)
> connect to a unix oracle 8 server on 2 dedicated private networks.
>
> At present, only one of the available networks is used in the
> TNSNAMES.ORA file on each of the appservers. Basically the raw IP
> address for the database(s) is specified. When I asked the developers
> if both addresses could be put there to give a redundant path to the
> database (if the 'primary' network failed) , the reply was that it
> slowed the performance, and was therefore removed.
>
> Sure enough, eventually there was a problem with the primary network,
> which was fairly easily solved by manually swapping the TNSNAMES.ORA
> on the appservers for one using the IP addresses of the other network.
> It still caused a bit of a stink though, so I had the idea of an
> automated check to swap the tnsnames on detection of a failure, but
> apparently this won't work as the file won't be read until oracle
> restarts, losing all the connections.
>
> Of course, if I had been involved in this project from the start I
> would have had something done to address this at the hardware or
> operating system level, but it's too late now. (typical!)
>
> Oracle is not my specialty. (I manage the NT in this situation) What I
> would like to ask you all is:
>
> 1) Can multiple network paths be specified in the TNSNAMES.ORA without
> causing any problems?
>
> 2) Is there a way to switch the TNSNAMES.ORA on the fly and maintain
> connections?
>
> 3) Can you suggest any other way in which this could be done? (without
> new hardware)
>
> Many thanks
>
> -Rob Stevens-
> NT System Manager
> Amey Comax (At DERA)
> rjstevens_at_dera.gov.uk
Received on Fri May 19 2000 - 00:00:00 CDT