LAN Pin Groups failover

[Krzysztof Zaleski]

Hi,

I am concerned about how traffic is distributed when we use static and dynamic pinning. Let's say, I have one chassis connected to only one FI. Two ports 1/1, 1/2 (etherchannel) go to one LAN switch, and next two ports 1/3, 1/4 (etherchannel) go to another LAN switch. Now, (1) how traffic from servers is distributed towards LAN if there is no static pinning (is hashing deterministic?), and (2) what if I have some servers pinned statically to one etherchannel, and other servers pinned to second etherchannel, and one LAN switch goes down, will traffic be moved to the other working etherchannel on the same FI?

Regards,

Krzysztof

[Marwan ALshawi]

mac-pinning:

forms ―subgroups based on uplinks, Each uplink (or bundled PC uplinks) is assigned a subgroup – up to 32 and VMs pinned to an uplink upon boot up

The MAC address of the VM will be used to select which link to use, same hashing used by vmware

If a failure occurs within the network, all the traffic pinned to the failed interface will automatically failover

Static Pinning

By default, VM traffic is pinned in a roundrobin fashion to the available uplinks

Static pining, select specific uplinks for specific traffic types based on need and availability requirements

You can use one links for SERVICE CONSOLE and another for VMOTION and you can keep active/standby failover for both traffic types

Wish this helps

If helpful rate

[Krzysztof Zaleski]

OK, so if I understand it right, if I use a deault dynamic pinning, traffic is autmaticaly protected in case of one link failure, and it is distributed among available uplinks according to a known algorighm. As the traffic distribution using dynamic pinning may not me exactly what I want, I can do static pinning to direct specific flows, and in case the uplink used in that static pinning does down, traffic will be re-pinned dynamicaly to other available uplinks, acting in a default fashion like there was no static pinning. And, if my uplink comes back up, the traffic will be re-pinned to that specific uplink, right?

[Marwan ALshawi]

Yes, also you need to take into consideration the type of bundling/grouping of the uplinks such as using etherchannle with or without vPC

fo rexample with N1K you can use LACP port channel, Allows for the VMs and VMKernal interfaces to utilise more than

one link for traffic Allows for fast vMotion and faster VM connectivity by using flow

based hashing BEST PRACTICE when connecting to upstream switches which are ―clustered upstream switches are ―clustered (vPC, VSS, VBS, Stack…)

if the upstream switch cannot be clustered use mac pining

good luck

if helpful Rate

[Krzysztof Zaleski]

Hm, something is not right. I cannot test any of those solutions so for now I can only get the teoretical knowledge, so I try to gid it up :-) You confirmed that when using static pinning, if uplink port goes down (doesn't matter if it is a single port or etherchannel, let's assume it just goes down), the traffic CAN be re-pinned dynamicaly to other uplink port on the same FI. But, I found a presentation from Cisco Live "BRKCOM-3002 Network Redundancy and Load Balancing Designs for UCS" where it is stated, that only dynamic pinning allows to re-pin on the same FI. Can someone confirm if it is true or not?

So, if I use static pinning with only one FI implemented, the only way to provide redundancy is to create two vNICs pinned to separate uplink ports and allow VMware use corresponding vmnic adapters as asctive/standby. Load balancing can be achieved by using the same method for other VMs, but different order of active/standby role on vmnics. Is my thinking right?

From the above presentation it looks like dynamic pinning is the best practice, but my concern is a load distribution. Using static pinning I have a better control over which server takes which path... What do you implement in your environments?

[Marwan ALshawi]

If it's same fi why u dont use etherchannel with static pining ?

Yes using active/standby and nic teaming kn esx level can work with two redundant fi !

Sent from Cisco Technical Support iPhone App

[Krzysztof Zaleski]

Etherchannel will not help me if my upstream LAN switch goes down :-) That's why I want two etherchannels to two switches. But, my design includes only one FI (don't ask why :-) ), so I wanted to provide 1) predictability in traffic flow 2) redundancy. Anyway, I keep finding different opinions on how pinning behaves, so I will have to test the solution once our UCS arrives.

[Marwan ALshawi]

Ok i c, in this case it will be redundancy to some extent not full because if the FI gose down then you will loose the connectivity

anyway try to define vNiCs ( diffrent port chnnels to the smae FI ) and present it to the ESXI host and in the VM level configure the NIC teaming to take care about the active/and standby links

havent done it like this but just try to present the NICs to the Vmware and from there try to configure the redundancy

by the way are you using the Palo card ? this one might help you interms of VNICs and NIC teaming on the VM level !!

good luck

[Marwan ALshawi]

by the way have you seen this one it might be useful if you did not

http://www.youtube.com/watch?v=Ru-1TztpfU8

good luck