I have all my vSwitches set up with primary and standby adapters. For the virtual machine switch, there are 4 adapters (two etherchannels) -2 nics primary, 2 standby. These are going to back to a pair of Cisco 6513's. IP has load balancing is set on the vswithch.
I test failover by pinging a guest and pulling cables. When it fails over to the standby, at most it'll miss one ping if that but when I fail it back to the primary by plugging those cables back in, it will consistantly miss 5 pings and then be back to normal.
Does anyone know of a setting that will enable the vSwitch to fail back faster?
euuhm this setup is a no go.
what will happen is when the first channel fails a link the first nic of the second channel will join in. in other words their will be two different channels active at the same time. create a 4 link channel instead or drop the channels and use load balancing on virtual port id.
My virtualisation blog:
I'm using two port channels as the primary and secondary nics go to different physical switches. If I fail one of the nics in that e-channel, at the most one ping will be dropped. Same failing to the standby nics, it is just failing back to the primary where there is the 5 ping delay.
The load balance policy is IP Hash.
Hmmm. That is not what I see happening: both nics in the first port channel have to fail before the the standby kicks in. If it did what you are saying, my testing would fail miserably.
It's an interesting Cisco article you posted on your blog Duncan (I read it briefly yesterday) - especially in regards to configuring an etherchannel and using Port ID on the vswitch/port group. We've only ever been able to use IP Hash.... will make some interesting testing on Monday :smileygrin:
Thanks Duncan - I'll pass this onto the netadmin of the site. If Cisco say's to do it - it's gotta be for a good reason. I'd say there will be a number of people out there now questioning their config after reading Cisco's pdf. But then I guess the other side of the arguement will be..... if it works, why change it?
Indeed Java, it amazed me that cisco prefers port id.... especially when you consider they can do cross stack etherchannels.
My virtualisation blog:
This is an interesting post, thanks Duncan. I just skimmed through it...had a question. On page 23 of the slides it shows a graph of ESX hosts, physical NICs and what looks to be virtual switches (pictured in the middle). Is it showing different hosts connecting to the same vSwitch or am I seeing it wrong? Is this a function of ESX 3.5 that I don't know about? vSwitches are limited to hosts I thought. I could be reading it wrong...
Duncan, right you are, thanks for pointing me to using esxtop to test. I still can't get it to drop any pings with 10 vm's behind the vswitch but I sure am seeing it start using one of the failover nics. I need to be going to redundant switches so I'll remove the etherchannels and go to port ID load balancing.