Forum Discussion

Stewart_76633's avatar
Stewart_76633
Icon for Nimbostratus rankNimbostratus
Sep 05, 2012

LTM vs GTM across DC's

Hi Guys,

 

I'm looking at putting in a couple of 1600's, one in each of our geographically diverse data centers. I hope to run these in an Active/Active HA pair where one side is responsible for some VIP's, and the other for other VIP's. If a fault existed in one DC I'd expect the HA to failover to the functioning side.

 

I was just wondering whether GTM is needed or whether LTM is sufficient to achieve the above?

 

We have high speed links between the sites (10gb) and low latency so I don't expect any issues in this regard.

 

Also, with the HA link, does the heartbeat network need to be on the same subnet, or can it be routable between subnets?

 

Cheers,

 

 

Stewart

 

config
design
  • Sec-Enabled_658's avatar
    Sec-Enabled_658
    Icon for Cirrostratus rankCirrostratus
    Sep 05, 2012

    Stewart,

     

     

    If you want true GLB, then you would want to use GTMs. GTMs respond to DNS queries based upon what resources are up in the various Datacenters it knows about. GTMs work hand in hand with LTM's to monitor resources in thier local Datacenters. If you didnt use GTM's you would need to have your LTMs monitoring resources across Datacenters which is not best practice.

     

     

    This is how it should work for your example:

     

    GTM in DC-A has a wideip pool containing two VIPs. One of the VIPs resides on the local LTM in DC-A and the other VIP resides on an LTM in DC-B. For DNS requests, the GTM responds with either of the two vips based upon RR. For whatever reason, if the LTM in DC-A has a monitor go down for its VIP, it would then tell the local GTM in DC-A, that its VIP is down. If as DNS request comes in for that wideip pool, the GTM would look at its wideip pool and see that the VIP in DC-A is down, but the VIP in DC-B is up still, so it would respond with the IP of the VIP in DC-B and not the VIP in DC-A. This will continue until the GTM in DC-A sees the resource become available again.

     

     

    As far as Heartbeat, are you using standalone GTMs in each Datacenter or are you using an actual HA pair of GTMs?

     

    If using an actual HA pair, then I would suggest using hardware failover cable, and not network failover.

     

     

    GTMs need to communicate to each over AND each LTM (All Datacenters) over tcp/22 and tcp/4353, and this CAN over a routed network.

     

    However,one thing to think about is, if you plan on using "Virtual Discovery" which I assume you want, you don't want the GTM talking to the LTMs through a NAT. http://support.f5.com/kb/en-us/solutions/public/9000/100/sol9138.html