EIGRP Load Balancing while running HSRP

[bstockslager]

Hi everyone,

I'm trying to configure load balancing on a network that is using 4 3750 L3 switches.  The network is comprised of two sites with two switches at each site running HSRP and the virtaul IP is the gateway for a server between the two switches.  The sites are connected by two independant 200meg ethernet connections (technically they are GigE limited to 200meg) p2p circuits.  The switches at site A are connected together via a GigE port and the server I have connects to each switch at its respective site using a GigE port and the same for site B.

The problem I'm running into is that I've noticed that one of the 200meg circuits is always idol unless the primary goes down, so I'd like to be able to load balance the traffic from site A to site B using both p2p circuits.  I think the problem that I'm running into has to do with the fact that HSRP is putting one of the links from the server to switch in standby unless the WAN link goes down and therefore traffic is not taking the "longer route" to the 2nd 200meg circuit.

The only traffic going across this network is only the two servers sending traffic to one another.

Does anyone have any ideas on how I can get load balancing to work or if it is a possiblity with this network topology?

Thanks in advanced,

Blake

[Richard Burts]

Blake

I am not sure that I fully understand your situation and your resulting question. But based on what I think I understand I have these observations.

You have 2 factors that make load balancing difficult to achieve:

- traffic between the sites is really traffic only between 2 servers. Since the default for load balancing is session based distribution of traffic it is difficult to balance when traffic comes from a single end point.

- if you are running a dynamic routing protocol then each switch will see 2 advertised paths to the destination subnet. One of the paths will be direct (essentially a single hop away) and the other path will be through the second switch at that site (essentially two hops away). This makes load balancing difficult.

One option you can consider to address the first difficulty is to specify cef load balancing to be per packet. This helps get around the single end point issue. Depending on the type of traffic per packet balancing may help or it may make performance worse (depending on the reaction of the application to out of order packets).

I can think of 2 possible options for the second difficulty. If you configure 2 equal static routes for the destination (one pointing to the direct connection and the other pointing to the other switch) you may get more load balancing. Or if you run EIGRP you might try the unequal cost load balancing that is an option with EIGRP.

HTH

Rick

[rich_foster]

Why don't you stack the 3750 switches and remove HSRP?  You can then rely on ECMP to load balance, i.e. the switch has two equal costs routes in its routing table and load balances accordingly.

[lgijssel]

Stacking would be a good option for sure but perhaps there are reasons why you are not using it?

To enable load balancing with EIGRP in this kind of situation there is the option to configure the variance.

Please check the link below:

http://www.cisco.com/en/US/tech/tk365/technologies_tech_note09186a008009437d.shtml

regards,

Leo

[rich_foster]

If you cannot stack the switches, maybe they are to far apart, here are some other options:

- Use GLBP rather than HSRP to load balance outbound traffic.

- Advertise a more specific route from one of the switches for half of the subnet.  For example if you have a /24, advertise a /25 from switch 2, as a result traffic will be split across the circuits.  This of course depends on your address allocations.

Also if you're given a 1gb bearer and your carrier polices at 200mb you should consider using the shaping feature:

http://www.cisco.com/en/US/docs/switches/lan/catalyst3750/software/release/12.2_52_se/configuration/guide/swqos.html#wp1200681

'In shaped mode, the egress queues are guaranteed a percentage of the bandwidth, and they are rate-limited to that amount. Shaped traffic does not use more than the allocated bandwidth even if the link is idle. Shaping provides a more even flow of traffic over time and reduces the peaks and valleys of bursty traffic.'

[bstockslager]

Thanks for the suggestions.

I think i'm going to try switching to GLBP instead of HSRP, that seems more suitable for what I need in my network.  The main issue that I'm worried about is achieving loadbalancing across the two switches when there's only one type of traffic going across the network, server A to server B and vise versa.  I'd like to keep the WAN connections themselves in seperate switches incase one fails there's still another path to take.

[bstockslager]

Well I tried using GLBP instead of HSRP but getting the same result, single path to host.  I think its because the LANs of each site only have 1 server sending 1 type of traffic across.

I'm beginning to think the only way to get use out of both circuits at the same time for this network is to put them both in the same switch and make a channel group to bundle them together.  The only reason I haven't is because this gives me a single point of failure.

Any futher thoughts?

Thanks a bunch for your help,

Blake

[rich_foster]

If you stack the switches, which is best option you can either:

- Have two layer 3 connections and load balance using EIGRP / CEF

- As you said create an LACP Etherchannel

With both of these options you can plug the circuits into different switches, and presumably dual home the servers.  The switch stack appears as one logical switch and has one control plane and you still have device resiliency.

As someone else mentioned you could use per-packet load balancing but this may introduce application layer problems, especially over the WAN.  Unfortunately if you create an Etherchannel you can still only load balance using src-dst-ip on the 3750 and not TCP or UDP, so traffic would only use one channel member:

http://www.cisco.com/en/US/docs/switches/lan/catalyst3750/software/release/12.2_55_se/commmand/reference/cli1.html#wp11908600

This is more of an CCIE R&S lab type solution, and probably won't work on the 3750 hardware platform, but here goes:  If you can differentiate based on application (despite only have one IP to work with), you may be able to use Policy Based Routing to route certain application flows across the other path.  You can apply a tracking (track) object to the PBR next hop and then track an EIGRP learnt route, interface or IPSLA probe in order to provide failover.

The best option is to use two IP addresses in each site!  You may be able to bodge it using a different static NAT on each side - but then you don't really want to do that....!

Please rate useful posts...