Multiple TCP/IP interface on the same subnet

[A.W.R]

Hi,

We have an HP Alpha Server with OpenVMS and TCP/IP. The system has 2 network interfaces one of 1Gbit/sec, and one of 100Mbit/sec.

They are both on the same subnet.

The 1Gig interface is used for backup traffic, and the 100Mbi interface is used for application traffic.

But we have found that all traffic goes through the 1Gbit/sec interface. We have had to disable the 1Gbit/sec interface.

Is there a way to control which interface is used for different targets/traffic?

Thanks
Andrew

[Steven Schweda]

> The 1Gig interface is used for backup
> traffic, and the 100Mbi interface is used
> for application traffic.

Says who? (Said how?)

> But we have found that all traffic goes
> through the 1Gbit/sec interface.

If you were the network software, and you had
a choice between a slow interface and a fast
interface, which one would you choose?

> Is there a way to control which interface
> is used for different targets/traffic?

Different subnets?

[Robert Badar]

Yes, there is a way. Based on routing settings, you can control traffic for different target addresses.

[A.W.R]

Hi,

The behaviour is just what we observed.

Is it possible to provide an example of the typical routing statements that would separate the traffic?

Thanks
Andrew

[Steven Schweda]

> The behaviour is just what we observed.

Why would it be any different?

> If you were the network software, and you had
> a choice between a slow interface and a fast
> interface, which one would you choose?

Well?

> Is it possible to provide an example of the
> typical routing statements that would
> separate the traffic?

The only ones I've been able to imagine would
involve using different subnets.

> They are both on the same subnet.

Are you willing to change that?

[Robert Badar]

Depending on your situation.
If everything is on the same subnet, then you can define only individual nodes to use different interface than default. In case of hosts on different subnet you can create routing rule for whole target subnet.

[Volker Halle · Accepted solution]

Andrew,

if you have multiple interfaces to the same subnet, TCPIP decides itself (based on a round-robin algorithm per outgoing IP session), which interface to use for OUTGOING traffic. This may also depend on your version of TCPIP.

INCOMING traffic will use the interface, whose destination address has been specified.

To specify a route for outgoing traffic, try:

$ TCPIP SET ROUTE /GATEWAY=

Volker.

[Hoff]

This is comparatively odd network design, which implies that there's some sort of staff-level communications confusion here, that there are network design details and requirements that have been omitted from the statement here, that the server or network hardware is unusually constrained, or that the networking folks here don't grok IP networking.

Personally, I'd be sending all the traffic through the GbE NIC under normal circumstances, and I'd be looking to establish QoS for the production traffic at the switches.

I'd be looking at deploying BGP or VLANs, depending on the traffic requirements.

If that switch-level work is not feasible here, then please subnet the network appropriately for your general requirements, or please manually set up static routes for the hosts. Of these two, establishing correct subnets is typically more appropriate.

The "fun" here tends to be with the NIC selected and used for the inbound traffic; with distributing this traffic across the available NICs, or targeting a specific NIC. That usually gets back to working within the switches/routers, or (less preferably) DNS-based load balancing. In the absence of network-level assistance, I'd want that to be the GbE NIC.

[Richard W Hunt]

Not sure about this, but if the two addresses are different and not declared as alternate paths, it should in theory be possible to hard-code everything by IP address. The problem is that you must keep any DNS from knowing about there being two ports into the same box.

I suspect that the problem comes about because some DNS somewhere knows that both x.x.x.x and x.x.x.y point to the same server. So... lie about it.

Machines these days are smart, but they are still just machines. As long as nobody TELLS the other machines "hey, here is this bloody fast I/F you can use for your app" then they won't use it.

Worst comes to worst, you can put a router-level ACL on the interfaces so that only your backup servers can see the fast I/F and then the other connection sources have to use what they can see - the 100 Mb port.

But then again, I have an issue of a different sort... if the same subnet allows that mixed traffic, why would you NOT just upgrade the 100 Mb card to a 1 GigE card and let EVERYONE have their fun at high speed? It's a one-time cost and a relatively small boost in the cost of the hardward maintenance contract for the faster card.

If your network team is worried about backup contention, you must be backing up continuously 24-hours per day. That's not good at so many levels that I don't know where to begin. This may be a confused policy issue, in which case the answer isn't networking cards or routes at all.

[A.W.R]

Thanks for your inputs - greatly appreciated.

Some time we find things that are not in an ideal situation, but we have to deal with it.