Inefficient GLBP scenario ?

Unanswered Question
Mar 9th, 2009

Normally I would be configuring HSRP or VRRP in a campus gateway backup scenario. Layer 2 at the access with VLANs spanning 2 Distributiion switches using HSRP or VRRP.

Would I be correct in assuming GLBP in this type of scenario can lead to inefficient traffic flows ? My thinking is that for L2 access layer there will be a link blocking back to the distribution layer. If the Distribution switch that is not root bridge for that VLAN is giving out the GLBP MAC address then my traffic goes from access layer down a L2 link to the first distribution switch and across the link to the other distribution switch to get routed. It may then possibly pass back over that link because the VLAN it is destined for may have its link blocked on the 2nd distribution switch.

With VRRP/HSRP since there is no load balancing I would just make sure that the active router follows the active root bridge for a VLAN then the first distribution switch I hit will do the routing and I won't be sent across the inter-distribution switch link unnecessarily.

Is this normal for GLBP ? Or can GLBP recognise the nearest router to the source of the traffic and give it the most appropriate MAC address as the gateway's address ?

I see that you can do weighting and tracking type configuration to help here but surely that's doing much that same as you would do with HSRP by ensuring one router always routes normally and another always routes if an interface connected to the first goes down.

Thanks, Stephen.

I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Loading.
Giuseppe Larosa Mon, 03/09/2009 - 05:40

Hello Stephen,

the target of GLBP is to provide effective load balancing providing a different MAC address (associated to a different AVF) in the ARP reply for the VIP.

Of course this means that some packets has to travel from access switch -- ds1 --- ds2.

>> VLAN it is destined for may have its link blocked on the 2nd distribution switch.

this cannot happen or the GLBP hello messages wouldn't arrive on the vlan.

But this shouldn't be a great concern in a real network unless your inteer-switch links are overused/full

the added latency is really small being just an additional L2 switch hop

Hope to help

Giuseppe

StevieOliver_2 Mon, 03/09/2009 - 05:52

Giuseppe

Perhaps I mis-described the line

>> VLAN it is destined for may have its link blocked on the 2nd distribution switch.

What I meant was that the link directly to the switch the end station is blocked meaning the traffic may have to be sent back to the 1st Distribution switch to be onward forwarded to the access layer. So the traffic goes -

Access switch -> 1st Dist switch -> 2nd Dist switch, gets routed onto another VLAN then traffic goes back to 1st Dist switch and out to Access switch.

The only load balancing being done here would be to spread the processing of the routing function over 2 processors. But additional switching is done to get the packet to it's designated router. I know it may be going across an under-utilised link between the dist switches with minimal latency but it's still inefficient.

Stephen.

Giuseppe Larosa Mon, 03/09/2009 - 06:14

Hello Stephen,

in inter vlan routing with GLBP and L2 access layer you can have:

0,

1,

2 inter-switch link transits

However, even with HSRP if the Active is not always the same switch for all vlans you can face some inter-switch traffic:

0,

1 inter switch link transits

if HSRP is active on all vlans on one device you have optimized the data path but no load balancing is in effect.

all uplinks to standby DS2 are idle (just signalling messages)

By the way to measure the changes in latency you would need a traffic generator that inserts a 32 bits timestamp inside the frames becuase it should be in the order of some tens of microseconds for GE links.

For example ping output in windows systems has a resolution of mseconds

We could say this is a reasonable tradeoff to use both DS instead of having one idle all the time.

Hope to help

Giuseppe

Mohamed Sobair Mon, 03/09/2009 - 07:44

Hi,

Adding to the above posts, GLBP provides GW loadbalancing per HOST. you will have loadbalanced of traffic on the Access based on the Active Virtual Forwarder MAC, but you would still have some kind of sub-optimal path for certain traffic since some traffic will be routed to the 2nd distribution layer which is not the Root Bridge.

With HSRP, VRRP based on your design , you would have normally the Root Bridge being the Virtual Router whether on the Access layer or the Distribution layer thus Having Optimal path of traffic flow without GW loadbalncing method for the Local hosts.

HTH

Mohamed

vivekmangla89 Mon, 09/15/2014 - 02:18

In my setup, two wan routers are connected to four WAN links, two from one ISP and remaining two from other ISP. I have configured by routers using glbp and a firewall and a few switches are sitting in between my clients and wan Routers. From firewall, all my traffic is going out via one router only and I don't think any load balancing is taking place. I have not tried trace from my clients. And I can't provide the glbp virtual ip as gateway in my client machines. So would the traffic from my clients to wan be load balanced by glbp ? From firewall, it's not being load balanced and using one router only.

Actions

This Discussion