Since all the vmware netcards are virtual there is no point juggling with them. Adding extra interface is just adding extra code to fail at random compared to simple netcard.
Under normal circumstances one would add following to /etc/modprobe.conf:
alias bond0 bonding
options bonding mode=6 miimon=100 (10^-3s)
mode=1 works better for legacy setups like hubs - one card in pool is up and when it goes down another goes up.
Then:
modprobe bonding
ifenslave bond0 eth0 eth1 eth2
Main Topics
Browse All Topics
by: warrenbucklesPosted on 2008-12-08 at 08:34:29ID: 23121673
I don't have experience with a setup like yours, but if you don't mind a bit of blind leading the blind, I have a few suggestions:
To grossly over-simplify, you have two systems that are interacting: the computer and the switches.
On the computer side you are trying to get the system to share traffic between links - load balance.
On the network side, you are trying to get the local switches to accept traffic from multiple MACs that are associated with a single IP.
On the computer side:
Have you tried disconnecting the connection to the card that is taking all the load?
If the other three are bonded correctly they should pick up and share the load.
On the Linksys unmanaged switch side:
This switch has fast ports, yes, but they can only be as fast as the processor that is managing them (or unmanaging them). In addition, you don't have any control over how the switch is prioritizing the ports.
[I wonder if STP (Spanning Tree Protocol) is active on your setup - it is designed to keep loops out of bridged switches by selecting one path between switches. I'm not sure if it is a problem here or if it can be dealt with (i.e., turned off) in your unmanaged switch.]
You might want to try bypassing the unmanaged switch, at least for the NFS part of the setup.
After that, you could try aggregating the ports on the managed switch - this is not a long-term fix, but it could point you toward a solution.
wb