Hi
My query is how do I configure bonding so that it can sense Uplink 1 failure in switch 1 and failover to switch 2 on which Uplink 2 is coming.
Here I have two HP L2 Procure Switches on which two network backend Uplink is connected.
I did the bonding configuration on the bases of arp request.
My configuration of modprobe.conf file was like this
alias bond0 bonding
options bond0 mode=0 arp_interval=100 arp_ip_target=192.168.1.1

the above options woked fine it sense uplink failure and failover the connectivity to switch 2 and vice versa. But this configuration had drawback that I was not able to take putty of this server although I was able to ping the 192.168.1.1 gateway from this server. I ran the tcpdump for figure out on both interface eth0 and eth1 I found first it sent ICMP packet from eth0 and second time from eth1.
Thus I changed mode=0 to mode=1 for dedicated interface I tried the failover testing of uplink again but this time it was successfully able to failover to eth1 to eth0 but not able to failback from eth0 to eth1. So how do I resolve this problem which can detect Network Uplink failiure of backend network.

Thanks & Regards
CJ

Can your switches be configured redundantly ? .. that would be my first choice. I'd let the switches deal with link failures upstream and use bonding to handle local link or switch failures

cheers

Do You think If I asked my Network Administrator to activate the STP and local loop on my local L3 switches would this will help..?

Thx & RGD
CJ

STP would most likely be on already.. but that doesn't mean the network is configured for redundancy, may be worth asking ...

No STP is not enable on backend network thus I am compelled to implement my solution. I tried it and check again the Uplink failure on my HP procure switches, now it is working fine with mode=1 also there was some mess up with network cables with switch while testing anyway... now it is ok. But my query is I have implemented on one server but what will happen if I implement this configuration on my 18 servers will this arp configuration creates flood on network. Say if I put 100 ms in arp_interval in modprobe.conf file.

If you run tcpdump or wireshark and filter on arp packets you should be able to estimate the amount of traffic generated by one machine with this configuration. Then simply multiply by 18 and decide whether this is an acceptable amount of overhead/traffic for your network.

cheers

Thanks I have checked from tcpdump but ironacily if i run tcpdump from node A it shows me another node B arp request to gateway and if i run from node B it shows me information of arp of node A. Why it is showing this why it don't show me arp information of the server on which i m sitting and executed tcpdump.

Rgd CJ

You should see any arp traffic for hosts that are in your broadcast domain, but you'll only see gratuitous arps when your link fails over so you may need to pull one of the cables. You will see normal arp requests when your machine is initiating a connection over the network to another machine and that machine has not been connected to recently ( arp cache entry has expired/been deleted ).

cheers