Load Balance of physical NIC

[sushil]

Hi,

I have a single upstream switch connecting to esx server on both the ports.On upstream switch ports say port2/3 and port3/3 allowed all vlans so that if one port fails other pick it up.That way I think failover will work for me.

On esx server these ports are added as plugges to vmnic1 and vmnic2.

I wanted to load balance the traffice so that can get aggregated bandwidth and failover as well.

I am sure it has to be done diffirently in this 1000v than physical.

Few questions;

1. Does this only command will work the traffic to be load balanced for me?

channel-group auto mode on mac-pinning

2. Does I need to create port channel/ether channel on uplink switch or port connecting from esx will pin it automatically.

So my requirement is to have aggregate the traffic as we do in physical switch.

Regards,

Sushil

[Robert Burns]

Sushil,

MAC Pinning does not "aggregate" bandwidth.  It operates the same way the vSwitch handles VM traffic by default which is "pinning" each virtual interface to one uplink.  Should that uplink fail, traffic will be re-pinned dynamically to another available uplink.  In the latest version of 1000v this pinning is distributed evenly across uplinks each time a new interface comes online.

To answer your quesitons. 

1. Yes - that's the only command you need in your uplink port profile. That easy!  I created a quick video to understand mac pinning a long time ago.  Though dated and done with an older version of 1000v, the concept is still the same: http://www.youtube.com/watch?v=4aHJtX3ZUZg

2. No. With MAC pinning you don't create any port channels on the upstream switches.  This is because neither MAC Pinning or the default vSwitch load balancing method aggregates the two links as one as mentioned previously.

If you want aggregate bandwith then you would use either mode "active" (aka LACP) or mode "on" which is a static port channel.  If you have a Cisco switch or one that at least support 802.3ad (LACP) then go with that.  It's preferred over static port channels.

Regards,

Robert

[sushil]

Robert,

Thanks a ton to you.

I am lamost done with the configurration.

95% is done except few features.

If you can help me on theses questions please;

1. I read that lacp is not possible if there is only one upstream switch.I gone throuh some videos where for LACP is enabled if upstream switches are clustered.In my scenario can I simply enable it where only one upstream is there?

2. The biggest confusion for me is the vlan i have mentioned for my control,packet,Data,vmotion etc.

Data is on vlan10-----> using a subnet 192.168.10.0/24 on upstream switch.All valns are trunked and created to nexus and vics-versa.

Packet is on vlan51-----> No ip address defined on this.Don't you think i should give it a ip address say 192.168.51.1 to it.

Same goes to rest of the vlan i.e control and vmotion and FT as well.Just merely giving name on upstream and nexus really not going to do anything.Confusion is there because vmware local tac asked me to just add vlan name and tag not any ip address but even they are not sure.

3. Finally What is the best practice for vmotion and FT.Should I use 3 uplink and use them as lacp(if possible on my example q1).In this way total bandwidth will be enough for me with redundancy.

If I use a dedicated NIC for vmotion and FT,Then it is great but if this particular nic goes down then I would get stuck on vmotion and FT.

Regards,

Sushil

[Robert Burns]

Answers inline.

Regards,

Robert

1. I read that lacp is not possible if there is only one upstream switch.I gone throuh some videos where for LACP is enabled if upstream switches are clustered.In my scenario can I simply enable it where only one upstream is there?

[Robert] Yes.  LACP does NOT require the clustering of switches.  You might be thinking of a VPC in this regard.  LACP port channels can be created on a single switch. LACP is an IEEE standard and usually available on most managed switches.

2. The biggest confusion for me is the vlan i have mentioned for my control,packet,Data,vmotion etc.

Data is on vlan10-----> using a subnet 192.168.10.0/24 on upstream switch.All valns are trunked and created to nexus and vics-versa.

Packet is on vlan51-----> No ip address defined on this.Don't you think i should give it a ip address say 192.168.51.1 to it.

Same goes to rest of the vlan i.e control and vmotion and FT as well.Just merely giving name on upstream and nexus really not going to do anything.Confusion is there because vmware local tac asked me to just add vlan name and tag not any ip address but even they are not sure.

[Robert] Since your control, packet, FT and vmotion VLANs are strictly Layer 2, they don't need a gateway.  They are not routed. Correct, just creating the VLAN and optionally giving it a name are all you need to do. 

3. Finally What is the best practice for vmotion and FT.Should I use 3 uplink and use them as lacp(if possible on my example q1).In this way total bandwidth will be enough for me with redundancy.

If I use a dedicated NIC for vmotion and FT,Then it is great but if this particular nic goes down then I would get stuck.

[Robert] I would suggest to aggregate as many NICs as you can using LACP.  As long as they're all using their own VLAN you can even throw in QoS into the mix when you're ready so no single network saturates your link.  Port channelling will give optimized usage of your links and provide the redundancy you want.  If you'd prefer to physically keep your VMotion/FT networks separate you can use a dedicated link - but as you know there's no backup if it fails.

Regards,

Sushil