GLBP host-dependent load-balancing - the right solution?

[Thomas Enchelmeier]

Hi all,
I have a question regarding the host-dependent GLBP load-balancing mechanism.

Situation:
We have two data center spanned via L2 extension.
In each data center we have two Catalyst 4500X coupeled to a VSS (one VSS in data center 1, one in data center 2).
In data center 1 the VSS will get the IPs: 172.16.45.252 (VLAN45) and 172.16.86.252 (VLAN86)
In data center 2 the VSS will get the IPs: 172.16.45.253 (VLAN45) and 172.16.86.253 (VLAN86)
The VIPs are 172.16.45.254 (VLAN45) and 172.16.86.254 (VLAN86)
The VSS should route the two VLANs.
To each VSS are some ESXi hosts connected (VLAN45), running some VMs (some also in VLAN45, some in VLAN86).
Target is, to have one default gateway configured for each VLAN on all ESXi and VMs for both data centers (the VIPs) .
The ESXi and VMs in data center 1 should always use the VSS in data center 1 as AVF.
The ESXi and VMs in data center 2 should always use the VSS in data center 2 as AVF.

Can GLBP in host-dependent mode do that? If yes, how to configure it, that it works like described?
If no, what alternatives do I have?

Thanks a lot for your help in advance.

Best regards
Thomas

[Francesco Molino · Accepted solution]

Hi

GLBP host-dependent isn't the right answer. It has 2 ways of load sharing and the one most used is based on arp. However, the algorithm used is round robin. That means some machines on DC2 can get DC1 as primary and this isn't what you want.

To achieve that on catalyst, it would be better to have 2 groups glbp or hsrp (1 for dc1 and 1 for dc2). This will guarantee to keep traffic staying in the DC and not crossing DC. However, your servers will have 2 gateways. I mean 1 VIP for all DC1 machines and 1 for DC2. 

Usually in such design I would prefer Nexus to leverage HSRP localization (FHRP). 

Sorry for that answer that don't give you what you were searching for.

Thanks

PS: Please don't forget to rate and mark as correct answer if this answered your question 

[Thomas Enchelmeier]

Hi,

thank you very much for your answer.

The missing puzzle piece for me was that I didn't know which algorhythm is used by the devices to decide which AVF will be used.
Since it is round robbin, like you stated, it can not be the right solution.

Best regards
Thomas

[Francesco Molino]

You're very welcome