LACP with Cisco Switches and NetApp VIFs

Hi Scott,

I just went through an exercise figuring out NetApp load balancing options late last year, and discovered the NetApp LACP support. One other dependency to note, for this configuration to supply both redundancy and load balancing across separate physical Cisco switches, you must use a Cisco StackWise stack.

a example for cisco stack with vlan

conf t
int port-channel 1
switchport access vlan 2
switchport mode access
exit
int ra gi 7/0/21, gi 8/0/21
switchport access vlan 2
switchport mode access
speed 1000
duplex full
no channel-group 1 mode on
channel-protocol lacp
channel-group 1 mode active
end
wr

My example:
4 port Channel, vith VLAN:s 860, 861 and 862

interface GigabitEthernet3/13
description NetApp link 1 of 4
no ip address
switchport
switchport trunk encapsulation dot1q
switchport trunk allowed vlan 860-862
switchport mode trunk
spanning-tree portfast trunk
channel-protocol lacp
channel-group 33 mode active
end

interface Port-channel33
description NetAPP_EthernetChannel
no ip address
switchport
switchport trunk encapsulation dot1q
switchport trunk allowed vlan 860-862
switchport mode trunk
lacp max-bundle 4
end

This works great on a single switch.

I’ve got (2) 3750g Cisco Switches connected to a 6500 series Cisco switch. 2 ports from my netapp go to each of the 3750 switches. I’d like to load balance between the 2 but it doesn’t work reliably.

I setup:
vif create lacp Switch1vif -b ip e0c e4c
vif create lacp Switch2vif -b ip e0d e4d
vif create multi vifCombo -b ip Switch1vif Switch2vif

This works to a degree. I can get data to come out of all 4 ports if I push it hard. But if I pull the cables from switch 1 it will take up to 30 seconds for switch 2 to takeover the connections. This is too long.

Hi Scott, is there also a solution how to configure ONTAP vif combined with LACP on a Juniper EX4200 switch? When trying to bundle 2 x 1Gb on the Juniper to the NetApp, I see the etherlinks go down now and then.

I have solution for EX4200, we bought EX4200 week ago, and after go through 1222 pages with lots of testing, we found the solution.

On netapp : vif create vifname lacp -b ip ……
will work with juniper lacp passive or active periode slow, no autonegotiation on physical port, link speed 1g
root@JS1# show interfaces ae0
description “test”;
aggregated-ether-options {
link-speed 1g;
lacp {
active;
periodic slow;
}
}
unit 0 {
family ethernet-switching;
}

On netapp: vif create vifname multi -b ip……
will work with juniper aggregated-ether-options , port mode access (no lacp)

it has been tested, it work pretty well….good luck

Hi,

We are currently setting up a netapp storage box and configuring a cisco 3750 stack.

I have created a port-channel and added two ports to this channel. Config below. Now if I set a continuous ping to the netapp and shut one port down it times out once then starts pinging again. Am I correct in thinking this shouldnt happen at all and that the ping should be continuous:

interface Port-channel3
description San 1
switchport access vlan 20
switchport trunk encapsulation dot1q
speed 100
duplex full

interface FastEthernet1/0/7
description 48a San Nic 3
switchport access vlan 20
switchport trunk encapsulation dot1q
speed 100
duplex full
channel-protocol lacp
channel-group 3 mode active
spanning-tree portfast
spanning-tree guard root
!
interface FastEthernet1/0/8
description 48a San Nic 4
switchport access vlan 20
switchport trunk encapsulation dot1q
speed 100
duplex full
channel-protocol lacp
channel-group 3 mode active
spanning-tree portfast
spanning-tree guard root

Shaun – About loosing ping

One thing that i noticed in your config – which may not makes difference – but worth checking
interface Port-channel3
description San 1
switchport access vlan 20 <<<<<<>>>
switchport trunk encapsulation dot1q <<<<>>>>

If using cisco, there’s no advantage in using lacp over standard etherchannel.

If you absolutely have to use lacp then use it in the ON mode, not active. Active mode starts negotiation which wastes precious seconds if you have a link failure. I can’t honestly call anything that doesn’t have sub-second failover an ‘HA’ system…

@sty: i don’t really agree. when having portchannels between switches, i found “active” to be better then “on” sometimes. First, true, active negotiation adds some delay. However, the goal of a portchannel is to keep the channel up and running at all times, when individual links fail/recover. I have seen extra packet loss when both sides are “on”: if one side is faster than the other one (ie. 6500 vs. 3750), with “on” , one side might start forwarding faster/before the other side is ready, resulting in 0.5 sec packet loss. When using “negotiate”, they will only start forwarding when both sides are ready and negotitation is successfull. result: 0 packet loss when a link bounces. Also, “on” is dangerous if the other side is “swapped or changed” and someone “forgets” to configure the interface. It will come up and the other side will start forwarding, often, resulting in 50% pkt loss.

Hello Everyone,

2 Questions in Regard:

1. In the vif0 configuration how do you enable jumbo frames on the Netapp?

2. Management:

EoA MANAGEMENT:
Originally I setup the N3300 with a separate non-routable data network on E0a and E0b, this is how I have done other installations. During one of our support sessions Charlie did some research and found out that in order to use the IBM System Manager and the FilerView you must assign a routable IP to E0a or E0b.

My expectation is, like that of other products, that ALL management traffic travel over a separate management network. In the case of the N3300 this is the BMC. Charlie informed me that only SSH works through the BMC, my experience was exactly that.

THE CONTRADICTION:
This is against Netapps published best practice of “separating IP based storage traffic from public IP network traffic by implementing separate physical network segments or VLAN segments”. Why is it that using the IBM/Netapp graphical management tools requires a contradiction of best practices? The separated data network is also meant to be a non-routable network, again this best practice contradicts the requirement of the management tools provided with the N3300.

QUESTION:
How can one manage the N3300 while separating the data network?

ANSWER:
The only answer I can come up with is separate VLANs on the same physical ports. However I do not agree

REFERENCE:
NETAPP TECHNICAL REPORT
NetApp and VMware Virtual Infrastructure 3
Storage Best Practices
http://www.netapp.com/us/library/technical-reports/tr-3428.html
page 18 for iSCSI, section 4.6
page 30 for NFS, section 5.5

Thank you,
Matthew