Agrregation on 4500

What you are asking to be done could theoretically be done from the server, but not to, unless you are using a load balancer. In short, its a software requirement, not hardware/network. You would need to use a virtual IP that load balances two separate actual IPs. You can only have one mac per IP, but multiple IPs per mac. Else layer two will not know how to forward the frames. The server could theoretically send out two NICs, but only one NIC will be the responsible mac for layer 3 coming in. Some kind of server side hsrp or vrrp type config also won't work as one server will always need to own the virtual, and they probably can't flip/flop fast enough to load balance equally. In short, load balancer!

Jamie, The load balancer solution might work if you are doing multiple backups at one time because the load balancer will balance based on separate TCP sessions. - but at that point, the load balancer itself could be the choke point. If you want to connect to both of the switches to the server for redundancy purposes, that's fine. But it won't increase "the bandwidth to the server" unless you do something like what Trendkill suggested - along with the caveat i mention above. If your goal is to simply increase bandwidth to the backup server (because you've verified is a link utilization issue) then I suggest you use Cisco's Port Aggregation Protocol and Etherchannel.

-JC

Thanks for the feedback...I am not sure about using loadbalancers at the moment. What the aim is to have two cables (one from each catalyst) and then have those two in an etherchannel configuration. If there is anyway of doing this using just the cat it would be great otherwise I might have to just stick the two cables on one of the catalysts and configure PAgP with Etherchannel.

If there is any other ideas would love to hear them.

Thanks.

Jamie

I do not see any way that you are going to form an etherchannel using ports from different switches. The commands for configuring an etherchannel do not have a place in the syntax of the command to specify another switch, only interfaces on that particular switch. The documentation for etherchannels for the 4500 series switches also does not include any references to the etherchannel being formed with any interfaces besides those within the switch itself.

It is technically impossible due to the one mac per IP deal. If the core arps for an IP, and receives one mac, it will use that one direction/port. If it received two, i guess it would send to both (guessing but not sure...never tried), and it would just eat up the bandwidth on both ports. What is being asked is to send some traffic to one mac, some to another, which would then require differing IP addresses and a load balancer because the session from the client is only going to be w/ one destination, not two. The only way you could do this without a LB, and presuming you know of the majority of clients, you could put one IP on one NIC, one IP on the other, and manually split your servers across the two IPs...configuring half to one, and half to the other. I'm pretty sure i read 'backup' above, so that may work, but if this is anything other than that and its random users (like a web app), this won't work.

Etherchannel cannot work across switches else it breaks the foundational rules of how l2/l3 work.

Hey all,

Thanks for the feedback- looks like a single switch it is then... ;-)

Jamie