GLBP Load Balancing Control on ISP uplink flapping

[abalasekaran]

i have GLBP configuration using round robin load balancing going to two different ISP through ( PR1 and PR2 )

If PR1 uplik is flapping / having high latency how do i control entire traffic over PR2 alone

[MHM Cisco World]

Can i see the topolgy?

[abalasekaran]

Thanks @MHM Cisco World  attached Topology for your reference,

[MHM Cisco World · Accepted solution]

https://cciethebeginning.wordpress.com/2008/09/06/glbp-gateway-load-balancing-protocol/

Glbp with track' if track up/down the weight is change.

[Joseph W. Doherty]

Unsure you can deal with your situation using just GLBP.

Indeed, GLBP can be configured to track an interface and change its active gateway status although it's more complex than a similar HSRP tracking setup.

In your case, besides flapping (interface up/down?), you've also mentioned latency!

Possibly (?) some SLA tracking might work, but SLA tracking is normally used to check reachability, not latency.

The one Cisco technology that I know addresses your issue is PfR.  (It can also dynamically LB and insure flows use the ISP that performs best per flow.  [It also doesn't even need Internet routes tables.])

[Joseph W. Doherty]

BTW, regarding GLBP tracking, I believe you first want to shift to weighted mode, setting equal weights.  (This should mimic your current round-robin mode.)

As weighted mode does proportional host distribution, based on weight, when your tracking adjusts weight, by default, it wouldn't shift all traffic, it would just shift some traffic.  Since you want to shift all traffic, you also need to set gateway activation/deactivation thresholds.  These should conform to whatever tracking weight change you decide upon.

Another change you might need to make is with the preempt setting.  Might only be a issue if you have more than 4 gateways.

If you need a suggested configuration, @MHM Cisco World might, if asked, lab something for you (as he's a lab wiz).

[abalasekaran]

I think GLBP with SLA tracking will fulfill our request partially, as i'm worried to prefer with PFR with current topology 

[Joseph W. Doherty]

PfR might work just fine with your current topology, but it does have a steeper learning curve than what GLBP will require (see me 2nd post).

Also, though, PfR might be a costly upgrade.  However, worthwhile knowing it exists, I believe.

Oh, if using the default FIFO queuing for ISP egress, you might consider trying egress fair-queuing.  Should be a standard feature; easy to configure.  Unfortunately, to obtain it's full benefit, ISPs would need to configure too, but no negative to only one side doing it.