[f-nsp] linux 802.3ad bonding with foundry switch

[Tamas Csillag]

Same experience here. Linux does support round-robin distribution of 
frames, but most switch vendors do not. We tried cisco/hp/smc/3com.

It kinda makes sense, as the RR method can produce out-of-order frame 
arrival, and the reassembly will use more CPU on the receiver side.
Guess it depends on the type of workload you actually want to balance.

Tamas


On 09/23/2011 10:19 AM, Jethro R Binks wrote:
> On Thu, 22 Sep 2011, Mike Allen wrote:
>
>> Off-hand, I think you want to go with link-agg active as opposed to
>> passive.  Not sure about the trunk commands, normally that would be for
>> setting up static trunks, but I don't remember the exact process in
>> older versions of switch and code (which I am guessing this is).  If you
>> could send model and version, that would probably help, the link
>> aggregation config has changed pretty dramatically over the years.
>
> Without wishing to detract from the original poster's question, I have a
> related one.
>
> I once looked at doing this with some non-Foundry kit and some Linux box a
> while ago: the client wanted to aggregate 4x 1Gb/s links for greater
> bandwidth (but not pay for 10Gb/s :).  Unfortunately, the connection over
> that link was pretty much from one client to one server (the latter was
> connected at 1Gb/s), and while the Linux client seemed to understand
> various modes of operation to spread traffic across several physical
> links, the other_vendor equipment didn't support it in the
> single-connection scenario: the connection was assigned a link based on
> various combinations of src/dst IP/mac, but only one link was selected for
> any particular combination.
>
> I'm a bit hazy on details now, but given the scenario of a single client
> trying to multiplex essentially one traffic stream to a single server over
> aggregated links, would that have been doable with a Foundry device?
> Something about "mode 0" at the Linux end rings a bell.
>
> Jethro.
>
>
>>
>> Mike
>> On Thu, Sep 22, 2011 at 7:51 PM, panicloop<panicloop at gmail.com>  wrote:
>>
>>> We did the connection test between cisco switch and linux server
>>> for "802.3ad bonding" mode.
>>>
>>> And we confirmed the following cisco's configuration works OK.
>>>
>>> --- begin ---
>>> !
>>> interface Port-channel1
>>>   switchport mode access
>>> !
>>> interface GigabitEthernet0/1
>>>   switchport mode access
>>>   channel-group 1 mode active
>>> !
>>> interface GigabitEthernet0/2
>>>   switchport mode access
>>>   channel-group 1 mode active
>>> !
>>> --- end ---
>>>
>>> And this time, we need to replace the cisco with the foundry switch.
>>> We did the foundry settings like following, but this doesn't works.
>>>
>>> --- begin ---
>>> !
>>> interface ethernet 9
>>>   link-aggregate passive
>>> !
>>> interface ethernet 10
>>>   link-aggregate passive
>>> !
>>> --- end ---
>>>
>>> Please anyone tell me how shoud I configure foundry switch.
>>> Should we do like following?
>>>
>>> --- begin ---
>>> trunk ethernet 9 to 10
>>> write memory
>>> trunk deploy
>>> --- end ---
>>>
>>> Sorry , I'm a linux guy and don't know much about switch.
>>> We need to connect the foundry switch and linux server using "802.3ad
>>> bonding mode".
>>> Thanks in advance.
>>>
>>> ______________________________**_________________
>>> foundry-nsp mailing list
>>> foundry-nsp at puck.nether.net
>>> http://puck.nether.net/**mailman/listinfo/foundry-nsp<http://puck.nether.net/mailman/listinfo/foundry-nsp>
>>>
>>
>
> .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .
> Jethro R Binks, Network Manager,
> Information Services Directorate, University Of Strathclyde, Glasgow, UK
>
> The University of Strathclyde is a charitable body, registered in
> Scotland, number SC015263.
> _______________________________________________
> foundry-nsp mailing list
> foundry-nsp at puck.nether.net
> http://puck.nether.net/mailman/listinfo/foundry-nsp