Channel Bonding: link aggregation across 2 switches

There are 2 switches on the data centre which is a switch managed by the
hosting company. Our servers have dual NICs in them and they are currently
configured using active-backup. Thus, only one slave is active at any time.

AFAIK, the switch is meshed together to provide redundant connections.

In this scenario, is it possible to have have link aggregation across 2
different switches that are meshed together ?

js wrote:
> There are 2 switches on the data centre which is a switch managed by the
> hosting company. Our servers have dual NICs in them and they are currently
> configured using active-backup. Thus, only one slave is active at any time.
>
> AFAIK, the switch is meshed together to provide redundant connections.
>
> In this scenario, is it possible to have have link aggregation across 2
> different switches that are meshed together ?

There are a variety of ways. I personally prefer doing this at layer 3
(IP layer), especially if your switches can also route at wire speed.

Basically, you assign each link an IP address and you assign the
machine a third IP address. You then advertise a route to the third IP
address over each link and arrange for traffic to be shared over
equal-cost links.

DS

js wrote:
> There are 2 switches on the data centre which is a switch managed by the
> hosting company. Our servers have dual NICs in them and they are currently
> configured using active-backup. Thus, only one slave is active at any time.
>
> AFAIK, the switch is meshed together to provide redundant connections.
>
> In this scenario, is it possible to have have link aggregation across 2
> different switches that are meshed together ?

That depends on the capabilities of the switches.

A $100 switch can't do anything of any real intelligence, but a $5000
one can.

Jeroen Geilman wrote:

> js wrote:
>> There are 2 switches on the data centre which is a switch managed by the
>> hosting company. Our servers have dual NICs in them and they are
>> currently configured using active-backup. Thus, only one slave is active
>> at any time.
>>
>> AFAIK, the switch is meshed together to provide redundant connections.
>>
>> In this scenario, is it possible to have have link aggregation across 2
>> different switches that are meshed together ?
>
> That depends on the capabilities of the switches.
>
> A $100 switch can't do anything of any real intelligence, but a $5000
> one can.

Assuming the switch does support it, is there anything special that needs to
be done on setting up the bonding interface ? Or is the work done the same
compared to setting up link aggregation of multiple NICs on just one
switch ?

js wrote:
> Jeroen Geilman wrote:
>
>> js wrote:
>>> There are 2 switches on the data centre which is a switch managed by the
>>> hosting company. Our servers have dual NICs in them and they are
>>> currently configured using active-backup. Thus, only one slave is active
>>> at any time.
>>>
>>> AFAIK, the switch is meshed together to provide redundant connections.
>>>
>>> In this scenario, is it possible to have have link aggregation across 2
>>> different switches that are meshed together ?
>> That depends on the capabilities of the switches.
>>
>> A $100 switch can't do anything of any real intelligence, but a $5000
>> one can.
>
> Assuming the switch does support it, is there anything special that needs to
> be done on setting up the bonding interface ? Or is the work done the same
> compared to setting up link aggregation of multiple NICs on just one
> switch ?

You just "bond" or "team" or "whatever" the NICs as you would normally -
whether two bonded NICs connect to one switch or to two different ones
should not matter as long as those switches communicate that fact to
each other.

You do need enough bandwidth *across* the mesh to support this, however.

If the link connecting the switches is slower than the resultant
bandwidth of 2 bonded NICs, the bottleneck will land squarely on that
link, and bonding the client NICs serves no purpose whatsoever.

Remember that end-to-end traffic will still need to go to *one*
endpoint, which is connected to either one switch, or the other.

Unless *all* systems are connected this way, in which case you could
(theoretically) turn a 1 gbit switching matrix into a 2 gbit one

So "the switch supports it" depends on two factors:
1. do they support trunking of bonded channels across the mesh link ?, and
2. are they themselves connected by sufficient bandwidth to make this
scheme even sensible ?

I know for a fact that most Cisco switches can do this.


HTH

J