3750 EtherChannel load-balance algorythm changed and loss of connectivity?

Unanswered Question
Jan 16th, 2008
User Badges:
  • Gold, 750 points or more

I recently changed the EtherChannel load-balance algorythm on some 3750 switches and lost connectivity. The switches are running the IP Services code and the EtherChannels are Layer-3 and spread over two switches in a stack (Port-Channel 1 is G1/0/1 & G2/0/1 and Port-Channel 2 is G1/0/2 & G2/0/2). They were configured with the default src-mac load-balance algorythm which obviously on a routed link has no load balancing effect since the source MAC never changes. I changed this to src-dst-ip and some hosts lost connectivity - i.e. I could ping from one VLAN to hosts somewhere in the rest of the network and from another VLAN I couldn't. The OSPF adjacencies stayed up on the stack so I am a bit lost as to why the connectivity failed. The load-balance decision is performed locally by the switch so why should I have lost connecivity?

Since this was a live network I removed the command to restore connectivity.


  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
skarundi Wed, 01/16/2008 - 07:59
User Badges:
  • Bronze, 100 points or more

perhaps the 2nd etherchannel member may have some Layer1/Layer2 issues. I would check the 2nd physical link that was actively used after the etherchannel load balancing algorithm was changed to be more effective.

andrew.butterworth Wed, 01/16/2008 - 08:48
User Badges:
  • Gold, 750 points or more

No, all the links are working fine as we tested the failover of links in the EtherChannel recently. I have also just had a look and no errors are shown (1000BaseSX fibre links).

I will probably change this again later on one of the switch stacks and dig a bit deeper. I have just re-read the documentation hoping there was an advisory or warning about changing the load-balancing algorythm when the channel was up but there is nothing.


andrew.butterworth Fri, 01/18/2008 - 02:19
User Badges:
  • Gold, 750 points or more

I managed to get a bit of time to play around with this yesterday and got it working, however I am not sure why what I did fixed it.

I made the changes again and had one of the server guys with me that could quickly detect any loss of connectivity. When I made the change immediately the connectivity issues were visible. To fix it I just shut the first port-channel down and then opened it up again, then repeated it on the 2nd once. Following these interface 'resets' the connectivity returned.

I did the same thing in another location and exactly the same behaviour was experienced - i.e. made the load-balance change and connectivity was affected, reset the etherchannels and connectivity was restored.

I can't find any notes on CCO regarding a warning on changing the etherchannel load-balance algorythm. Has anyone else seen this and has an explanation?



This Discussion