FailsafeIP with multiple networks

[Jeremy Begg]

Hi, I am configuring some new Integriity blades which are running OpenVMS IA64 V8.4 with TCP/IP Services V5.7 ECO 3.  Each blade has a number of NICs, four of which are critical to this question:

EWA0 and EWB0 are on network 1 (IP network 172.22.66.x)

EWC0 and EWD0 are on network 2 (IP network 172.22.208.x)

Each NIC is connected to a different switch via Virtual Connect.

There are servers are at two sites and there is (supposedly) redundancy in the paths between the sites, so that if (say) the path connected to EWA0 fails, the path connected to EWB0 will still be active.  However there is no automatic failover (at the physical inter-site link level) from one path to the other.

I'm trying to work out how to get IP failover happening between EWA0 and EWB0, and between EWC0 and EWD0.  It seems to me there are two options, FailsafeIP and LAN Failover.  However I suspect LAN Failover won't be entirely successful since it relies upon there being a problem at the level of the NIC or the link between the NIC and its local switch.  I can easily imagine scenarios in which the NIC and its switch are happily talking to each other but the path to the remote side is lost.

If I set up FailsafeIP, can there be two separate FailsafeIP "sets" on the same box?  For example, if EWA0 stops receiving network packets for some reason, I don't want its IP address being assigned to EWC0 or EWD0.  (Doing so would achieve nothing, since they're on a different IP subnet.)

If FailsafeIP can do this, how many IP addresses do I need? It's unclear from the documentation if it's one per NIC, or one per pair of NICs (EWA/B, EWC/D).

Thanks,

Jeremy Begg

[Duncan Morris · Accepted solution]

Jeremy,

you could get away with just 2 ip addresses in this configuration.

Set up ip address 1 as active on EWA0, and as standby on EWB0. You would then configure ip address 2 as active on EWC0 and standby on EWD0.

The addresses only failover to NICs which have been configured for that specific address, not just to any free NIC.

Configuration options:

   0  -  Set The Target Node (Current Node: hostx)

   1  -  LE0 Menu (LLA0: TwistedPair 100mbps)
   2  -  192.168.10.18/24    hostx_serv1           Configured,Active
   3  -  192.168.10.41/24    hostx_web             Configured,Active-Standby

   4  -  LE1 Menu (LLB0: TwistedPair 100mbps)
   5  -  192.168.10.41/24    hostx_web             Configured,Active
   6  -  192.168.10.18/24    hostx_serv1           Configured,Active-Standby

   7  -  IE0 Menu (EIA0: TwistedPair 100mbps)

   8  -  IE1 Menu (EIB0: TwistedPair 100mbps)

   9  -  IE2 Menu (EIC0: TwistedPair 100mbps)

  10  -  IE3 Menu (EID0: TwistedPair 100mbps)

In the example above, we have EIA0 and EIB0 configured for LAN failover as LLA, and EIC0 and EID0 configured for LAN failover as LLB. There are 2 services offered on the system, serv1 and web.

The ip address for serv1 is active on LE0 (==LLA0) and on standby on LE1 (==LLB0). Conversely the ip address for the web service i active on LE1, but standby on LE0.

[Andy Bustamante]

You can configure TCPIP failSAFE as proposed.  for an excellent review of options, see The OpenVMS Technical Journal articale  http://h71000.www7.hp.com/openvms/journal/v2/articles/tcpip.pdf by one of the TCPIP Engineers (Hi Matt).

A recommended configuration would be to assigned each interface an address.  Add a failSAFE address to each pair of interfaces and provide this as the service address.  This allows management interface access to each interface. 

Since failover is determined by traffic and lack of traffic, if you have any other traffic to the interface (local subnet traffic) you may not have the fail over addresss shifted. 

[Jeremy Begg]

Hi Duncan,

Thanks for that example, it's very close to what I want.  One difference, when I ran TCPIP$CONFIG.COM I didn't see that list of menu options (no LL0 devices) but re-reading the doco suggested that FailsafeIP requires (at a minimum) the same IP address on two interfaces.  So that's what I did, and it seems to work.

Andy - thanks for the feedback.  I had already read Matt's excellent article but it emphasis the desirablilty of having multiple IP addresses per interface, which isn't the way I want to go.  (I can understand why a site would want to do it Matt's way, but we have other methods for gaining access to the system if both interfaces in a FailsafeIP set fail.)

Jeremy Begg