GLBP or stick with HSRP?

[niro]

All of our servers are plugged into one of two 6509 switches. The 6509's are connected to each other with Layer 2 trunks, and we're running HSRP on all the vlans for redundancy (Default gateway on the servers is the HSRP address). I've been thinking about switching over to GLBP, but I'm wondering if it would really provide any benefit over HSRP in this scenario?

[Edison Ortiz]

GLBP provides redundancy just like HSRP but will also provide load-balancing unlike HSRP.

If you want to load-balance the traffic, then GLBP would be the recommended solution

HTH,

__

Edison.

[Amit Singh]

In the present scenario, switching over to GLBP will not do much of the difference and will work the same way as the HSRP.

-amit singh

[Joseph W. Doherty]

Real benefit depends.

GLBP often improves gateway router load balancing. With HSRP, one gateway router accepts all the (to the) gateway traffic, with GLBP multiple routers share the (to the) gateway traffic. (With HSRP, load can be distributed by alternating gateways for multiple subnets.)

If the gateway routers, have equal cost routed uplinks, equal cost routing protocols often won't share them. (From the initial gateway router, its peer's path includes the cost to the peer, making that path more costly.) Since GLBP splits the load to the gateways, often the load is split the same across the routed uplinks.

From the L2 edge, and using a triangle STP loop, usually one uplink is blocked, so GLBP would still have all the traffic flow up the same active link; some traffic would cross over to the other GLBP gateway(s) via L2 trunk. (Can happen with HSRP too.)

GLBP can be especially nice in a branch with a single subnet and with only two WAN routers. For your situation, you'll need to decide if any of the above makes it appear worthwhile to switch.

PS:

If you do switch, note that GLBP uses different virtual MACs from HSRP, so hosts "wonder" what happened to the gateway until the next ARP.

[niro]

Thanks Joseph.

Our servers are plugged directly into one of the 6509s, and the VLAN they plug into run HSRP, and the 6509 is the default gateway for these servers (the HSRP address).

So what I was wondering was, for the servers that are plugged into the secondary 6509 which has the standby vlan, all the traffic from the servers would have to go through the Layer 2 trunk to hit the Primary HSRP address, right? Will running GLBP help prevent that? From the sound of it, even with GLBP, some servers would still send traffic destined to the default gateway through the Layer 2 link, depending on which mac address they received, is that right? If that's the case I will just stick with HSRP.

[Joseph W. Doherty]

I'm a bit confused. Is the standby vlan the same subnet as the primary? If the same subnet, the server brings it's second link on-line (or moves its MAC) if it loses the first? (Some variant of NIC teaming?)

If what you're describing, the primary server path is lost, but that 6500 is still active, your concern is that HSRP doesn't move so all the traffic is drawn to the primary 6500 from the secondary across the L2 trunk. Yes, GLBP would help, but not what you want, I suspect.

GLBP will likely already be drawing half the traffic to the secondary (what you're trying to avoid altogether), but with the loss of the server's primary path, GLBP will like draw half the traffic to the primary from the secondary (better than HSRP's drawing all).

[niro]

Well...some servers are plugged into the primary 6509 and some servers are plugged into the secondary 6509...they all currently have only one connection, and I'm talking about a single VLAN, the servers are just load balanced accross the two 6509s. I'm not trying to keep traffic off of the secondary (secondary is not the correct term in this case, I guess), I'm just trying to keep traffic from crossing the trunk unnecesarily...and utilizing both 6500's more efficiently.

[Amit Singh]

Hi Niro,

You shouldnt worry about the traffic for inter-vlan routing.Although in HSRP the other 6509 RP will be in standby mode but it will still do the inter-vlan routing for you locally on the same switch. The inter-vlan routing traffic will not pass via the trunk link to first 6509 for inter-vlan routing. The trunk link will only be used for intra-vlan traffic. It will only be used for inter-vlan traffic when you dont have the same vlan gateway on both the switches.

Hope this makes sense to you.

HTH,

-amit singh

[niro]

Yea I'm not so much worried about inter-vlan routing, because like you said, as long as both hosts are on the same switch it will stay there...I'm more concerned with routing routing to other vlans.

[dbroder]

One more thing I'd throw out the is GLBP's ability to use group numbers above 256 (GLBP can go up to 4096 I believe). I match my GLBP group #s to match my VLAN #s - makes my code self-documenting.

Darren.

[lamav]

Niro

One thing that I dont see mentioned on here is that GLBP uses asymmetric routing by design. So, if you are using some sort of server load balancing scheme, like IOS-based SLB or a CSM, dont use GLBP. You will have intermittent connectivity problems....

HTH

Victor

[caplinktech]

Another thing I do not see mentioned here is the lack of redundancy at all in the design.

Unless I missed something, HSRP/GLBP is not really going to help much as if one of your switches fail, whatever systems are attached to that switch will be offline.

HSRP/GLBP only work if you redundant uplinks that allow physically allow the traffic to be transported to both switches. You seem to indicate that each server is only connected to 1 or the another switch. If the switch fails, how do you expect the server to reach the internet?

As Joseph hinted at to make HSRP/GLBP useful in any way you would need to have some type of NIC teaming on the servers and redundant uplinks, otherwise for the most part HSRP is just wasting cpu cycles.

[niro]

Thanks guys.

I'm aware that if one switch fails all the devices on that switch will not work (since they have no other connectivity). That's not my concern at this time, alot of our more critical servers are redundant at the application level (core 1 fails, servers on core 1 lose connectivity, applications running on servers on core 2 pick up the slack). Routers and switches are mostly plugged into both switches for redundancy using HSRP.

I guess what I'm going to end up doing is switching one or two vlans to GLBP and see how it affects traffic flow across the trunk link.

Thanks for your guys input.