Link bonding across multiple switches, am I right this way?

Question

After reading https://wiki.linuxfoundation.org/networking/bonding at "Maximum Throughput in a Multiple Switch Topology", I wonder whether I can link to Host A with single LAN port after a 802.3ad switcher like this:

                   +-----------+
                   |  Host A   | 
                   +-+---+---+-+
                     |   |   |
            +--------+   |   +---------+
            |            |             |
     +------+---+  +-----+----+  +-----+----+
     | Switch A |  | Switch B |  | Switch C |
     +-+----+---+  +-+---+----+  +-+---+----+
       |    |        |   |         |   |
   +---+    |        |   |         |   |
   |  +-----|--------+   |         |   |
   |  |  +--|------------|---------+   |
   |  |  |  |            |             |
   |  |  |  +--------+   |   +---------+
   |  |  |           |   |   |
+--+--+--+-+       +-+---+---+-+
| Switch D |       |  Host B   | 
+-----+----+       +-----------+
      |
+-----+-----+
|  Host C   |
+-----------+

Switch A,B,C are ordinary unmanaged switchers, Switch D is a managed one with 802.3ad support.

Host A and Host B are both configured as BONDING_OPTS="mode=4 xmit_hash_policy=layer3+4"

On Switch D, after set the 3 upper ports being bonded, will I be able to access Host A from Host C ?

BONDING_OPTS="mode=4" will not work either with or without Switcher D and Host C, BONDING_OPTS="mode=0" should be working.

Is there some way to reduce broadcasting since Switcher D will link other switches.

TDurden · Accepted Answer · 2016-12-16T16:17:44.493

2

I redrafted your diagram to make it a little easier to understand what is going on.

Scenario A is how you presented your problem.

Since you are using unmanaged, dumb switches, the diagram can be simplified and collapsed into Scenario B. You are trying to establish two Link Aggregation Groups (LAG) (802.3 ad), comprised of three links to Switch D from Host A and Host B. These LAG links are represented by ovals in the new diagram.

In theory, your traffic should pass between all hosts. HOWEVER, because you are directly connecting Host A and Host B to unmanaged switches, your LAG links may not form correctly. The resulting connectivity between Host A or Host B and Switch D would be unpredictable.

Additional Caveats

LAG groups in Cisco IOS are called EtherChannels. EtherChannels require dedicated interfaces per EtherChannel groups (Port Channel) formed. Even if Switches A, B and C had 802.3ad support and were Cisco IOS, the EtherChannels would not form.

Spreading LAGs across physical devices is not typically supported by most vendors. Cisco implements this feature with Virtual Switching System (VSS) and a Virtual Port Channel (VPC) is formed. https://www.cisco.com/c/en/us/products/collateral/switches/catalyst-6500-virtual-switching-system-1440/prod_qas0900aecd806ed74b.html

edited Dec 16 '16 at 16:17

answered Dec 16 '16 at 15:28

TDurden

246
1
3
6

The original "Multiple Switch Topology" without `switcher D` and `Host C`, does that work with `mode=4` ? Is it possible to access them from `Host C` in other topology ? – Galaxy Dec 16 '16 at 16:10
Unfortunately, no. The original 802.3ad specification has a limitation of each LAG to one logical switch. See this link and look under the section for limitations. https://en.wikipedia.org/wiki/Link_aggregation – TDurden Dec 16 '16 at 16:25
So, what about `mode=0` on `Host A` and `Host B`, and all 4 switchers are dumb, will that work ? I want to reduce some broadcasting, seems this will be the same under one dumb switcher. – Galaxy Dec 16 '16 at 16:34
Mode=0 is still bounded by 802.3ad limitations. All modes are based on 802.3ad. Mode=0 specifies traffic utilization of the LAG in a round-robin fashion. – TDurden Dec 16 '16 at 16:44
In order to reduce your broadcast domains you must introduce a broadcast boundary such as a VLAN or Layer 3 device such as a router. – TDurden Dec 16 '16 at 16:46
In which case can "Maximum Throughput in a Multiple Switch Topology" in that webpage work ? According to your opinions, it seems that is totally wrong. – Galaxy Dec 16 '16 at 16:48
If the above answer, is satisfactory, please accept it as a "Solution". This will mark the question as "Solved" to close out the topic. I am still happy to try to answer follow up questions in the comments as best as I can. – TDurden Dec 16 '16 at 16:49
Let us [continue this discussion in chat](http://chat.stackexchange.com/rooms/50204/discussion-between-galaxy-and-tdurden). – Galaxy Dec 16 '16 at 16:51
I disagree with some details in your answer. If the bond is connected to three separate IOS-style EtherChannels, the bond will still negotiate, but it will negotiate three separate Aggregators and **only one Aggregator can be active at a time** so effectively only one slave and switch will be used. You are correct that what OP wants is switches that support VPC or other MLAG. – suprjami Jan 03 '17 at 11:33

Link bonding across multiple switches, am I right this way?

1 Answers1