Load Sharing, four switches, two WAN Links - Design

[stephen.stack]

Hi All,

Hope someone can help me out here. Need to get a design complete and i need some ideas

The basic topology - Two Sites - 2 Cisco 3560Xs at each site. Each pair of switches is trunked

using etherchannel. I've got 2X 1GigE links between each site, and the idea is to terminate each

link on a single switch. See diagram below.

How can i acheive full load sharing accross these links dynamically??

I was hoping to use GLBP and OSPF load sharing over the 1GigE links

but sadly the 3560s do not support GLBP (oversight on my part).

Either a layer2 or Layer3 solution is acceptable.

Thanks in advance

Stephen

[Jon Marshall]

Stephen

L3 between the sites is preferable if you don't need the same vlan in both sites as -

1) you eliminate STP between the sites so a L2 issue in one site is isolated from the other site

2) you can use both links as equal costs path if you use a dynamic routing protocol such as OSPF/EIGRP

As for load-balancing internally, assuming the clients are connected to the 3560 switches then you would have to look at MHSRP because as you say GLBP is not supported. MHSRP allows you to configure multiple groups under the same LAN interfaces so you would in effect have 2 VIPs for each pair of L3 interfaces. And you make one router active for one of the VIPs and the other router active for the other VIP.

There are 2  big drawbacks with MHSRP (which is a large part of the reason GLBP was invented)  -

1)  you must manually configure the clients with the correct VIP. This can still be done through DHCP but you must obviously configure half the scope with one VIP and the other half with the other VIP.

2) because it is not dynamically allocated as in GLBP it is not a precise load-sharing setup as you may at any one time have more clients with one VIP than the other.

See this doc for links on configuring MHSRP -

Load sharing with HSRP

Jon

[stephen.stack]

Hi Jon,

Thanks for your opinions on this.

I agreee that L3 is the way to go, and i intend using OSPF for equal cost load balancing.

As you point out my issue is around internal redundancy/ load sharing. MSHRP is not really going to cut it i think.

Too much management over head. Shame GLBP is not available - can see why not though.

I'm trying to think of anything i can do with GRE tunnels or the likes, but nothing is comming to mind.

Regards

Stephen

[Jon Marshall]

Stephen

What exactly are you connecting into the 3560 switches at each site and how are they connected. Bear in mind that if it is just clients then HSRP is limited anyway as you can only connect them as single honed devices so if the switch fails HSRP gives you nothing in terms of clients connected to that switch.

I only ask because if they are single honed clients in one of the sites then you could actually run both intersite connections to just one of the switches and make that the L3 switch and then have the other switch connected via an uplink to the first and acting purely as L2. That way you would indeed be able to utilise both links but with the obvious downside that you now have a single point of failure.

Another option may be to use 3750 switches instead and use cross stack etherchannel between the 2 switches.

Other than that if the switches host the client/servers directly there is little you can do.

Jon

[stephen.stack]

Thanks Jon,

Your help is appreciated, as always

My options are now a bit limited, and i am sutck with the hardware i have.

The ultimate goal is to use both links as you're aware.

I am now thinking about using both links as L3 links and utilising EIGRPs

unequal cost load balancing. (Perhaps even trunk the links as use SVIs for the

EIGRP adjacencies)

Any opinons on this?

Thanks a mil

Stephen

[Jon Marshall]

Stephen

Always glad to try and help.

To be honest i'm not sure this is any better a solution than MHSRP. You are going to get a lot of interconnect traffic between the 2 switches at each site and although i have never used unequal cost load balancing there are a few people on these forums who have reported problems with it.

Unfortunately I don't think there is a particularly "clean" solution to your problem because your switches don't support GLBP and you cannot run an etherchannel across 2 switches as you can with 3750 switches.

What exactly do you have connected to the switches in each site and how are they connected ie. dual honed/singly honed ?

Jon