Sign in with
Sign up | Sign in
Your question

Considering UDP/TCP ports in 802.3ad Link Aggregation

Last response: in Networking
Share
Anonymous
July 4, 2004 1:00:20 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

Hi all,

I was wondering how common it was in practise to determine conversations
as per 802.3ad when aggregating Ethernet links based on layer 3 and layer
4 information. That is,

(1) would commonly used switches do that at all and
(2) do folks use that feature then?

Btw, I do understand about the respective client-server, server-server,
etc. scenarios. What I would much appreciate your feedback on is the
practical aspect of whether or not the feature is widely spread throughout
today's product landscape, and if you use it.

I cannot afford buying a bunch of commonly used switches and test it
myself, so that is why I hope to be enlightened here. ;-)


Thanks and tc,
/Tom.

--
Thomas Bahls | "Wege entstehen dadurch, daß man sie geht."
Greifswald, Germany | -- Franz Kafka

ICQ #119230485 +++ PGP Key 0x6E70B6AE +++ http://solitaryhiker.net/
Anonymous
July 10, 2004 10:25:49 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

The primary purpose for these options is to give you control of traffic
distribution over the aggregated links. Most admins would like traffic to
be randomly distributed across the bundle. MAC addresses provide the
highest degree of randomness because in general, you leave the address set
to the factory value. The MAC distribution methods provided by the switch
vendors often use the last byte(s) for the algorithm. IP addresses are
controlled by you and may have a deployment which is less random from an
algorithm point of view. For example, maybe odd value IP addresses are in
one location and evens in another (wild speculation here to demonstrate a
point) This may cause an imbalance for the distribution. Adding Port
values (L4) assists in the randomization.

A reason that you might use L3 for distribution is that some routers use
L3 methods. You may want to use the same algorithm as the routers for
consistancy, but this is not mandatory.

As to which switches do what, it depends :-)It is a vendor feature
implementation. Even MAC
distribution methods can vary by vendor. Some may use the last byte, some
more.
Some the source, destination or both.

Hope this provides some food for thought.

KevinOn Sun, 04 Jul 2004 09:00:20 +0200, Thomas Bahls <tbahls@arcor.de>
wrote:

> Hi all,
>
> I was wondering how common it was in practise to determine conversations
> as per 802.3ad when aggregating Ethernet links based on layer 3 and layer
> 4 information. That is,
>
> (1) would commonly used switches do that at all and
> (2) do folks use that feature then?
>
> Btw, I do understand about the respective client-server, server-server,
> etc. scenarios. What I would much appreciate your feedback on is the
> practical aspect of whether or not the feature is widely spread
> throughout
> today's product landscape, and if you use it.
>
> I cannot afford buying a bunch of commonly used switches and test it
> myself, so that is why I hope to be enlightened here. ;-)
>
>
> Thanks and tc,
> /Tom.
>



--
Kevin Hamilton
Co-author: Cisco Press CCIE Development Series: Cisco LAN Switching
CCSI, RCSI, RCIE, RCNP, JNCIS
July 12, 2004 1:11:37 AM

Archived from groups: comp.dcom.lans.ethernet (More info?)

"Thomas Bahls" <tbahls@arcor.de> wrote in message
news:40e7aaf8$0$21305$9b4e6d93@newsread2.arcor-online.net...
> Hi all,
>
> I was wondering how common it was in practise to determine conversations
> as per 802.3ad when aggregating Ethernet links based on layer 3 and layer
> 4 information. That is,
>
> (1) would commonly used switches do that at all and
> (2) do folks use that feature then?

like most things - it depends.

the algorithm is often (always?) driven by hardware caches - so if you have
a layer 2 only switch, it may be "blind" to higher layer addresses, so layer
3 is not practical
or the traffic may not be IP - difficult to balance that by IP address, or
TCP / UDP socket.....

layer 3 IP balancing is sometimes needed (e.g. if there is a single main
server at 1 end of a balanced link, there may not be many MAC addresses
involved) and is normally supported in a layer 3 switch - the better ones
let you choose by source or destination address, or both.

not heard of using socket numbers - i guess it is feasible, and would work
with some traffic patterns.
>
> Btw, I do understand about the respective client-server, server-server,
> etc. scenarios. What I would much appreciate your feedback on is the
> practical aspect of whether or not the feature is widely spread throughout
> today's product landscape, and if you use it.

Common practice on switches from Cisco, Nortel, Foundry, 3Com, Enterasys -
usually for Gigabit links on recent network designs....
>
> I cannot afford buying a bunch of commonly used switches and test it
> myself, so that is why I hope to be enlightened here. ;-)
>
>
> Thanks and tc,
> /Tom.
>
> --
> Thomas Bahls | "Wege entstehen dadurch, daß man sie geht."
> Greifswald, Germany | -- Franz Kafka
>
> ICQ #119230485 +++ PGP Key 0x6E70B6AE +++ http://solitaryhiker.net/
--
Regards

Stephen Hope - return address needs fewer xxs
!