linux bonding: ping statistic = 50% loss
 

Hi,

I have a problem here implementing an effective bonding interface.
I follow the instruction from http://www.cyberciti.biz/tips/linux-...interface.html

The difference is that, each interface eth0 and eth1 were connected to different adsl modem, meaning they are not going to the same destination, instead each heading for individual destination modem. The target we tried to achieve here is so that with the setup, we can aggregate the bandwidth of the two adsl links.
eg eth0 -----> 192.168.1.10
eth1 --------> 192.168.1.100


At the end, i successfully get the 3 interfaces up when doing less/proc/net/bonding/bond0 i can see
Output:

Bonding Mode: load balancing (round-robin)
MII Status: up
MII Polling Interval (ms): 0
Up Delay (ms): 0
Down Delay (ms): 0

Slave Interface: eth0
MII Status: up
Link Failure Count: 0
Permanent HW addr: 00:0c:29:c6:be:59

Slave Interface: eth1
MII Status: up
Link Failure Count: 0
Permanent HW addr: 00:0c:29:c6:be:63

But, the problem is when i tried to ping either 192.168.1.10 or 192.168.1.100 , the ping statistics always result in 50% loss.
Say if i ping 192.168.1.10, the iptraf display that each interface eth0 and eth1 took turn to ping, but then the outcome of the ping is always 50%...

I am using red hat enterprise 4 and 3comm network card. I have tried to use different mode such as mode 0 and mode 6 but the outcome still the same...
Any idea what else can I fine tune so the ping result will have no loss?

thanks a lot for taking time helping!

regards
y

This seems more like an attempt at BGP...

You can't 'balance' between 2 providers; How can you expect the routing of packets to work !?

My guess is that your ping issue is this: The two real interfaces probably don't route between each other (I don't think you'd want that anyway). So when you try to ping one interface or the other, since the pings are being routed back and forth between the two interfaces, you're going to fail half the time, which is what you're seeing.

But I think the point mtimbro is making is this: You can't start a connection via one route and have the packets for that connection just go back and forth out of two providers. What you'd want to do is somehow (and I don't know enough about linux networking to know how you'd do this with iptables, or if it can be done), is to basically have each connection have its own interface.

One a connection is established, all the future packets for that connection are marked as being "related" to the established connection. What you'd want to do is somehow make sure all related packets go out the same interface as the packet that established the connection.

If that can't be done, another possibility is simply to figure out some way to split the routing 'space' into two. That is, say, route all traffic to [0-127].x.x.x through one adapter and all traffic to [128-255].x.x.x through the other adapter. Although I probably wouldn't quite split it that way since it won't give you even splitting.

Actually, what probably would give you pretty fair splitting is if you did it based on the 3rd or 4th (probably the 3rd) number. That is:

x.x.[0-127].x and x.x.[128-255].x