Connecting Server with Teaming

[engineer_msu]

Dear Experts, I have one doubt and need clarification on this.

I am connecting a server with four LAN connections to Cisco Switch. Below are two secnario, i am confused with.

1. At the server I am doing teaming but at the switch side just connecting all the four cables in access vlan - Connectivity is working with redundancy

2. At the server I am doing teaming and at switch side I have created port-channel and configured all the four ports in this ether channel - connectivity is working with redundancy.

Please let me know Solution 1 is correct or do I need to use solution 2, to get redundancy as well as higher bandwidth.

Thanks in advance.

[Hitesh Vinzoda · Accepted solution]

As far as my practical experience with HP team they work in following modes.

1. NFT Network fault tolerance

2. NFT with preference

3. Transmit load balancing with NFT

4. TLC with NFT and prefrence order

5. Switch assisted Load balancing and FT ( Etherchannel LACP comes in the play)

6. 803.ad dynamic and FT

Basic teaming can be done by only having access port configuration on the switches ( server may connect to different switches, no issues).. You may create etherchannel only when you choose for switch assisted load balancing.

Thanks

Hitesh

[engineer_msu]

So its clear that there is no need to configure ether channel at switch side for teaming to work. Please clear if I configure ether channel at switch side, do I need to use protocol LACP or just configure with default vlaue.

[Steve Fuller · Accepted solution]

Again this will depend upon the server teaming mode. The IEEE 802.3ad standard specifies two methods of operation: static and dynamic. Static is simply that where interfaces are assigned to a link aggregate and there is no interaction between the link partner. The dynamic mode is when the Link Aggregation Control Protocol (LCAP) comes into play and requires interaction and negotiation between the switch and the server.

In the HP methods mentioned above, the Switch-Assisted Load Balancing (SLB) does not use LACP so on the network switch you would configure channel-group <group> mode on. For the 802.3ad dynamic method then you would configure channel-group <group> mode [active|passive].

Regards

[Steve Fuller]

Hi,

This will really depend upon the server OS and the NIC teaming / bonding mechanism that's being used. You can increase both resilience and capacity using either your scenario 1 or your scenario 2 providing the server bonding configuration is setup correctly.

The scenario 1 is what is sometimes referred to as "switch independent" teaming. That is, there is no configuration required on the network other than placing all the ports that the server NICs connect to into the same VLAN.

Depending upon the server OS, the load balancing method in use could be active/standby or active/active. For example in Linux there is a method called active-backup (mode=1)  which uses only one NIC for sending and receiving traffic. One of the other NICs in the "bond" is used if, and only if, the active NIC fails. By way of contrast there is mode=5 (balance-tlb) whereby the server will send traffic on more than one NIC, and mode=6 (balance-alb) whereby the server will send and receive on more than one NIC

The scenario 2 you talk about is called "switch dependent" teaming and will use link aggregation mechanism such as the IEEE 802.3ad standard. When using this method then you need to configure a port-channel interface on the network and assign the physical ports as members of that port-channel. In this method traffic is able to use all NICs in both the send and receive directions. One drawback to this method is that, according to the IEEE standard, there can be only two "link partners". In the past this would have meant one server connecting to one switch which meant no resilience to switch failure. Today, with the advent of technolgies such as VSS on some Catalyst switches, vPC on the Nexus switches etc., you can connect the NICs across two physical switches and so get the required resilience as well.

The scenario 1 is very simple, requires no real interaction between network and server admins, and is generally easy to troubleshoot. The scenario 2 is a little more involved, and while network engineers are generally happy with port-channels and link aggregation, I've known server admins to avoid it as they feel it's complex.

If you work in an environment where you have VSS or vPC, and the network and server admins work well together, then I'd favour the link aggregation methods over an active/backup method.

Take a look at the CentOS Bonding Interfaces page for Linux, or the HP ProLiant network adapter teaming white paper for background reading.

Regards