VMware Solutions Discussions

Active/Passive and Aggregate Bonding of Network

CAPPPER12
5,464 Views

We just purchased a FAS2240A-2 and a DS4243 Shelf.

We have a vision of creating an aggregate and bonding 2 of the GigE ports on one head and 2 of the GigE ports on the other.  (see Attached)

I am not sure if NetApp can bond the GigE ports into an aggregate.  Is this possible?  Also, does NetApp care if the links are routed to different switches or do they have to go the same switch for an aggregate?

As for the iSCSI connections we would like to do either an Active/Passive or Active/Active connection.  Not sure the pros and cons of either.

Also, not sure how the data is routed.  Is the one line inert until needed?  Are both active and traffic load balanced?  Does it choose best available route, etc?  How does NetApp handle Active/Active or Active/Passive network links?

Thank you in advance.

1 ACCEPTED SOLUTION

yannb
5,464 Views

Hello,

Yes, you can create an aggregated interface from two physical ports. It can be configured as simple aggregate or "dynamic" LACP links.

Unless you are going to switches from the same chassis or themselves aggregated using proprietary protocols so that they are seen as a single switch, no, you can't go to different switches because the network will not "understand" paquets coming from different interfaces for a single MAC address.

As for iSCSI and ESX, there is an excellent technical report on the subject that you must read because :

  1. It is not trivial
  2. It contains massive amount of informations about link aggregation, ESX iSCSI implementation, or NFS

http://www.netapp.com/us/system/pdf-reader.aspx?m=tr-3749.pdf&cc=us

View solution in original post

3 REPLIES 3

yannb
5,465 Views

Hello,

Yes, you can create an aggregated interface from two physical ports. It can be configured as simple aggregate or "dynamic" LACP links.

Unless you are going to switches from the same chassis or themselves aggregated using proprietary protocols so that they are seen as a single switch, no, you can't go to different switches because the network will not "understand" paquets coming from different interfaces for a single MAC address.

As for iSCSI and ESX, there is an excellent technical report on the subject that you must read because :

  1. It is not trivial
  2. It contains massive amount of informations about link aggregation, ESX iSCSI implementation, or NFS

http://www.netapp.com/us/system/pdf-reader.aspx?m=tr-3749.pdf&cc=us

CAPPPER12
5,464 Views

Yann,
Thank you very much for this. I will take a look at the technical report.

Just to clarify, are you saying the intelligence on the NetApp, when an aggregate is created, bonds the two interfaces and presents them as one MAC? I know the aggregate presents the link as one IP but I want to clarify that the interfaces are or are not presented to the network as a single MAC. The only reason I am asking is because we are currently running a similar configuration on Citrix Xen Appliances where we have aggregated interfaces connected to different core switches and the functionality has no issues. I am just trying to figure out if NetApp functions differently and we know for a fact that it will not work.

Thank you again

yannb
5,464 Views

My answer was completely misleading and confusing.

What the switch will not understand is having a single IP address coming from two different MAC addresses. And that situation is not comparable to what happens in Xen or Vmware when aggregating links : in that case, you actually get one physical port assigned to a VM, so it is (by default) always the same IP going out (and in) to the same interface. I don't know about Xen, but in VMware, you can tweak this behavior and apply a policy based on IP hash so that a new interface gets a chance to be picked up each time the couple source-destination ip changed. BUT in that case you have to pay attention to your switch settings, and be sure to be connected to an aggregated link on the switch side (so it is not good to go to two different switches, unless they share some kind of switch level aggregation), and I'm pretty sure the same applies to Xen. You can look at it that way : Xen or Vmware are binding VM's to ports, so that there is no MAC/IP confusion. ONTAP has one IP you are distribution to two ports, hence the confusion if the switch is not aware of that.

Public