MCSG support with 100BaseT and 1000BaseSX

[Darren Gibbs]

Current MCSG environment consists of a 2-node cluster with an N4000 and RP7400. Planning to upgrade the RP7400 to a Superdome with 1000Base-SX and no 100BaseT. If we do this, are there any MCSG software or support issues with running a failover node with 100BaseT and a primary node with 1000Base-SX? Specifically, will the heartbeat and package IP's be able to commmunicate properly between the two with the extra network hops?

[Stephen Doud]

Hi Darren,

Local Switching Restrictions
Local switching (between two network cards on the same machine) is only supported between network interface cards of the same type. More specifically, local switch works between NICs that have the same network stack. For example, all Ethernet cards use the same network stack, so they can be used in any combination for local switching. This means that network interface cards such as 10Base-T, 100Base-T, 100VGAnyLan, 1000Base-T and 1000Base-SX can be combined for local switching. An example of combinations of cards that cannot be used for local switching are Ethernet cards with FDDI, Token ring, ATM or Fibre Channel cards, since all of these cards use different network stacks.

Remote Switching Restrictions
Like local switching, remote switching (between different machines) is only supported between network interface cards of the same type. More specifically, remote switching only works between cards which have the same network stack. All Ethernet cards use the same network stack, so they can be used in any combination for remote switching. This means that network interface cards such as 10Base-T, 100Base-T, 100VGAnyLan and 1000Base-T can be combined for remote switching. For example remote switching between a 100Base-T Ethernet card on one node and a 1000Base-T Ethernet card on another node is supported. Remote switching between a 100Base-T Ethernet card on one node and a FDDI card on another node is NOT supported since these cards use different network stacks. Other combinations of cards that CANNOT be combined for remote switching are Ethernet with FDDI, Ethernet with Token Ring, FDDI with ATM, etc.
It is important to remember that if network cards of different speeds are combined for remote switching, it is essential to ensure that the slowest card is capable of handling the bandwidth requirements for your package applications. If this is not the case, then your package application network performance may degrade severely when the package is running on the host with the slowest network card.


-S.

[Sridhar Bhaskarla]

Hi,

If you are using local lan failover, then you will need to make sure that the lan cards are of the same type. Otherwise you are good to go.

Also I would configure atleast two heartbeat lans seperate from the package lans.

-Sri

[Darren Gibbs]

Since it seems this will work. I plan to order 4 1000BaseSX cards for the primary servers. One will be dedicated to the package, one will be for local lan failover of the package IP, one will be dedicated for heartbeat and one will be for a backup VLAN. So in all I'll have 4 heartbeats.

The failover node will stay as it is today with 3 100BaseT cards and 1 1000BaseSX card. The 100BaseT will be for the package IP, one for standby of package IP, one for heartbeat and the 1000BaseSX fiber for backup VLAN. Again, a total of 4 heartbeats.

[Sridhar Bhaskarla]

Hi Darren,

I would suggest the following if have the network infrastructure.

3-1000Sx cards and one 4 Port 100BT lan card.

Use two ports of the 4 Port card for heartbeats.

Out of three SX cards, one is for the primary and the other for failover. The third Sx card is for backup as well heartbeat. This way if the entire 4-port card fails, you still have a heartbeat going on.

You probably will have to buy 1 four port and 1 1000-Sx card if you don't have them already for the failover server. Make the 1000-Sx card for backup as well heartbeat to match the primary server.

-Sri