load balancing or sharing on ethernet
Phil Howard
Suppose you have two groups of ethernet nodes. You have a big switch at
each group. Now suppose the total aggregate traffic between the two is
greater than 100 megabit. Can you connect TWO 100BaseT connections between
them to get greater speed? What technology is necessary to do this? Will
spanning tree recognize both paths and make full use of them?
--
-- *-----------------------------* Phil Howard KA9WGN * --
-- | Inturnet, Inc. | Engineering Operations Manager | --
-- | Business Internet Solutions | eng at intur.net | --
-- *-----------------------------* philh at intur.net * --
Robert Morella
lets back-date some terminology here....a switch is very similar in
function to a bridge...transparent bridging makes use of the spanning
tree alogorithm vs. Source Route Bridging.
If you just hooked two 100Bt connections to inter-connect two
switches, transparent bridging would go into a 'loop' because the
alternate paths might be learned by two ports on one switch...in other
words a duplicate mac address would appear in two port tables on the
switch and it would be very confused.
Spanning tree algorithm would come to the rescue and prune duplicate
paths and create a 'loop free' path. Thus this would block traffic
on duplicate-loop ports....your second port would go dead.
it's possible to do source-route bridging on ethernet (ug), but it
wouldn't really buy you anything since either of the two ports on the
same switch could not assigned a higher prioriity than one another.
The problem with the dual cable approach under srb is that srb
discards alternate routes and srb broadcast searches suck up
bandwidth...especially if SNA is being used.
There are a couple ways to get there:
1) Cisco has a proprietary technology called 'etherchannel' which
allows aggregation of this sort.
2) You could do something with VLANs. If your switch supports
cross-chassis VLANs, I believe it would be possible to segment each
switch into two VLANs common with two VLANs on the other switch.
Now you could put a router between VLANs...or a layer3 switch..or
maybe somehow make some common VLAN.....hmmm need to ponder this more.
3) buy two switches with a gig-ethernet uplinks.
Gary MacIsaac
Phil Howard wrote:
> Suppose you have two groups of ethernet nodes. You have a big switch at
> each group. Now suppose the total aggregate traffic between the two is
> greater than 100 megabit. Can you connect TWO 100BaseT connections between
> them to get greater speed? What technology is necessary to do this? Will
> spanning tree recognize both paths and make full use of them?
>
Hi,
You probably want to look at switches which can implement port trunking between
them. I think this will be a proprietary solution. The HP 800T is one product
which will do this:
http://www.hp.com/rnd/products/switches/800t/800t.htm
Cheers, Gary.
Neil Schroeder
Phil Howard (NOS...@rembrandt.intur.net) wrote:
: Suppose you have two groups of ethernet nodes. You have a big switch at
: each group. Now suppose the total aggregate traffic between the two is
: greater than 100 megabit. Can you connect TWO 100BaseT connections between
: them to get greater speed? What technology is necessary to do this? Will
: spanning tree recognize both paths and make full use of them?
All Cisco switches I believe support FastEtherchannel, but on a cheaper
and higher performance note Foundry Networks stackable products do
100BaseTX trunking as well as gigabit trunking... but you have to have
some proprietary solution currently.
--
Neil Schroeder ne...@mindspring.net
Senior Network Engineer +1 404 815 0770x2560
MindSpring Enterprises, Network Engineering www.mindspring.com
Eric Pylko
In article <69ehn9$4...@camel15.mindspring.com>,
Neil Schroeder <ne...@excelsior.netops.mindspring.net> wrote:
>Phil Howard (NOS...@rembrandt.intur.net) wrote:
>: Suppose you have two groups of ethernet nodes. You have a big switch at
>: each group. Now suppose the total aggregate traffic between the two is
>: greater than 100 megabit. Can you connect TWO 100BaseT connections between
>: them to get greater speed? What technology is necessary to do this? Will
>: spanning tree recognize both paths and make full use of them?
>
>All Cisco switches I believe support FastEtherchannel, but on a cheaper
>and higher performance note Foundry Networks stackable products do
>100BaseTX trunking as well as gigabit trunking... but you have to have
>some proprietary solution currently.
I would say most cisco switches support etherchannel; for the Catalyst line,
only certain cards support etherchannel.
Also, if you were running vlans like somebody said, you can have multiple
connections between switches by setting the vlan priority on different
trunks. I'm pretty sure that is documented on their web page.
-Eric
--
Eric Pylko Phone: (716) 253-1611
Network Engineer Pager: (716) 975-1792
py...@kodak.com Fax: (716) 726-7283
phil-k
Just finished designing LAN and project managing with Cabletron Switches.
They will provide port trunking
for 3 connections. It doesnt matter whether these are 10 or 100
connections. Problem is no duplex.
--
Phil Kurek
Kurek Management Group
Strategic Network Consulting
Eric Pylko <py...@blue.pixel> wrote in article
<69fp70$n4c$1...@thetimes.pixel.kodak.com>...
William Welch
Yes there is something available called MLT or multi-link trunking.
Where you would trunk more than one port via two switches.
Most schemes are proprietary at this point so you will not be able to
mix and match vendor switches.
Bill Welch
Bay Networks
Phil Howard wrote:
>
> Suppose you have two groups of ethernet nodes. You have a big switch at
> each group. Now suppose the total aggregate traffic between the two is
> greater than 100 megabit. Can you connect TWO 100BaseT connections between
> them to get greater speed? What technology is necessary to do this? Will
> spanning tree recognize both paths and make full use of them?
>
Anoop Ghanwani
Robert Morella (rmor...@mindspring.com) wrote:
: NOS...@rembrandt.intur.net (Phil Howard) wrote:
: >Suppose you have two groups of ethernet nodes. You have a big switch at
: >each group. Now suppose the total aggregate traffic between the two is
: >greater than 100 megabit. Can you connect TWO 100BaseT connections between
: >them to get greater speed? What technology is necessary to do this? Will
: >spanning tree recognize both paths and make full use of them?
: lets back-date some terminology here....a switch is very similar in
: function to a bridge...transparent bridging makes use of the spanning
: tree alogorithm vs. Source Route Bridging.
: If you just hooked two 100Bt connections to inter-connect two
: switches, transparent bridging would go into a 'loop' because the
: alternate paths might be learned by two ports on one switch...in other
: words a duplicate mac address would appear in two port tables on the
: switch and it would be very confused.
: Spanning tree algorithm would come to the rescue and prune duplicate
: paths and create a 'loop free' path. Thus this would block traffic
: on duplicate-loop ports....your second port would go dead.
[stuff deleted]
: 2) You could do something with VLANs. If your switch supports
: cross-chassis VLANs, I believe it would be possible to segment each
: switch into two VLANs common with two VLANs on the other switch.
: Now you could put a router between VLANs...or a layer3 switch..or
: maybe somehow make some common VLAN.....hmmm need to ponder this more.
A proprietary VLAN solution may be possible. IEEE 802.1Q VLANs operate
over a single spanning tree. So, you'd still have the spanning tree
problem pointed out above.
The IEEE 802 committee is currently working on a standard for
trunking which would allow one to group two or more links
in such a way that they look like a single switch port.
-Anoop
Magnus Homann
a...@eden-gardens.ee.duke.edu (Anoop Ghanwani) writes:
> The IEEE 802 committee is currently working on a standard for
> trunking which would allow one to group two or more links
> in such a way that they look like a single switch port.
Interesting. What happens when one of the trunked lines goes down?
Will there be dynamic loadsharing?
What number is the committe?
Yours,
--
Magnus Homann Email: d0a...@dtek.chalmers.se
URL : http://www.dtek.chalmers.se/DCIG/d0asta.html
The Climbing Archive!: http://www.dtek.chalmers.se/Climbing/index.html
Anoop Ghanwani
Magnus Homann (d0a...@palver.dtek.chalmers.se) wrote:
: a...@eden-gardens.ee.duke.edu (Anoop Ghanwani) writes:
: > The IEEE 802 committee is currently working on a standard for
: > trunking which would allow one to group two or more links
: > in such a way that they look like a single switch port.
: Interesting. What happens when one of the trunked lines goes down?
: Will there be dynamic loadsharing?
: What number is the committe?
There was some debate about whether it should be done
in the 802.1 (bridging) or 802.3 (Ethernet) working group.
Right now it looks more likely to be done in 802.3.
-Anoop
Tom Turnbull
Have you tried setting both switches to full-duplex if this is available on your
switch?
Tom.
Phil Howard wrote:
> Suppose you have two groups of ethernet nodes. You have a big switch at
> each group. Now suppose the total aggregate traffic between the two is
> greater than 100 megabit. Can you connect TWO 100BaseT connections between
> them to get greater speed? What technology is necessary to do this? Will
> spanning tree recognize both paths and make full use of them?
>
Rich Seifert
In article <69ihra$lpk$1...@news.duke.edu>, a...@eden-gardens.ee.duke.edu (Anoop
Ghanwani) wrote:
> The IEEE 802 committee is currently working on a standard for
> trunking which would allow one to group two or more links
> in such a way that they look like a single switch port.
>
Well, it hasn't gotten quite that far yet. The IEEE 802.3 Working Group
approved the formation of a *study group* to look at starting a formal
standards activity for trunking. This group will meet for the first time in
Bellevue WA in February. (I will be there.) It is expected that a formal
standards group will be proposed at the March IEEE plenary meeting in
Irvine CA.
--
Rich Seifert Networks and Communications Consulting
sei...@netcom.com 21885 Bear Creek Way
(408) 395-5700 Los Gatos, CA 95030
(408) 395-1966 FAX
"... specialists in Local Area Networks and Data Communications systems"
Look for my upcoming book, "Gigabit Ethernet: Technology and Applications
for High-Speed LANs", from Addison-Wesley! (Available April 1998)
Phil Howard
On 15 Jan 1998 01:38:50 GMT Anoop Ghanwani (a...@eden-gardens.ee.duke.edu) wrote:
| Magnus Homann (d0a...@palver.dtek.chalmers.se) wrote:
| : a...@eden-gardens.ee.duke.edu (Anoop Ghanwani) writes:
|
| : > in such a way that they look like a single switch port.
|
| : Will there be dynamic loadsharing?
|
| : What number is the committe?
|
| There was some debate about whether it should be done
| in the 802.1 (bridging) or 802.3 (Ethernet) working group.
| Right now it looks more likely to be done in 802.3.
How complicated can this be?
Once a switch has determined that port 1 and port 2 actually do connect
to the same box, it should be able to send some frames one way and some
the other way. Even integrating this with VLAN should be no problem as
long as there is a way to determine some node id of the other box.
OTOH, I've seen many protocols that get more complicated than they need
to be just to suit a committee of people with interests in making sure
some piece of their pet technology gets used.
--
-- *-----------------------------* Phil Howard KA9WGN * --
-- | Inturnet, Inc. | Director of Internet Services | --
Phil Howard
On Tue, 13 Jan 1998 09:09:29 -0500 William Welch (wwe...@baynetworks.com) wrote:
| Yes there is something available called MLT or multi-link trunking.
| Where you would trunk more than one port via two switches.
|
| Most schemes are proprietary at this point so you will not be able to
| mix and match vendor switches.
That sounds like what they are doing is either enveloping the frames in
something proprietary, or using some new proprietary thing to recognize
that the two links go to the same place.
I would hope that spanning tree would have offered the means to recognize
that port 1 and port 2 are in fact the same node, and that from then on,
it just sends frames over one or the other port. There wouldn't be any
need to frame them differently. All that it would seem to me to be needed
is to know that both ports truly are the same box on the other end.
Phil Howard
I don't have the switch, yet. I was waiting to see what switches might
actually do this, if any.
On Thu, 15 Jan 1998 12:47:39 +0000 Tom Turnbull (Tom.Tu...@durham.ac.uk) wrote:
| Have you tried setting both switches to full-duplex if this is available on your
| switch?
| Tom.
| Phil Howard wrote:
| > Suppose you have two groups of ethernet nodes. You have a big switch at
| > each group. Now suppose the total aggregate traffic between the two is
| > greater than 100 megabit. Can you connect TWO 100BaseT connections between
| > them to get greater speed? What technology is necessary to do this? Will
| > spanning tree recognize both paths and make full use of them?
| >
| > --
| > -- *-----------------------------* Phil Howard KA9WGN * --
| > -- | Inturnet, Inc. | Engineering Operations Manager | --
Daniel J McDonald
Phil Howard wrote:
>
> On 15 Jan 1998 01:38:50 GMT Anoop Ghanwani (a...@eden-gardens.ee.duke.edu) wrote:
> | Magnus Homann (d0a...@palver.dtek.chalmers.se) wrote:
> | : a...@eden-gardens.ee.duke.edu (Anoop Ghanwani) writes:
> |
> | : > The IEEE 802 committee is currently working on a standard for
> | : > trunking which would allow one to group two or more links
> | : > in such a way that they look like a single switch port.
> |
> | : Interesting. What happens when one of the trunked lines goes down?
> | : Will there be dynamic loadsharing?
> |
> | : What number is the committe?
> |
> | There was some debate about whether it should be done
> | in the 802.1 (bridging) or 802.3 (Ethernet) working group.
> | Right now it looks more likely to be done in 802.3.
>
> How complicated can this be?
>
> Once a switch has determined that port 1 and port 2 actually do connect
> to the same box, it should be able to send some frames one way and some
> the other way. Even integrating this with VLAN should be no problem as
> long as there is a way to determine some node id of the other box.
And you have to make certain that the frame is not flooded back to the
original box over a different link, or over a link that goes to another
switch and then to the first one, so that bridge loops aren't created.
To do that effectively, you really need to know the entire topology and
calculate which switch the mac address is originally associated with,
and only radiate away from the core, and also do so in a loop free
fashion. In other words, you have to duplicate layer three routing
without the logical abstraction layer of layer three addressing. Sounds
pretty hard to me!
--
Daniel J. McDonald Intel Corporation
daniel_j...@mail.intel.com Oregon Site IT
1.503.677.5542 (voice) CCIE # 2495
Disclaimer: Not speaking for Intel Corporation
Alan Deikman
Magnus Homann wrote:
>
> a...@eden-gardens.ee.duke.edu (Anoop Ghanwani) writes:
>
> > The IEEE 802 committee is currently working on a standard for
> > trunking which would allow one to group two or more links
> > in such a way that they look like a single switch port.
>
> Interesting. What happens when one of the trunked lines goes down?
> Will there be dynamic loadsharing?
You throw all the traffic normally slated to the downed line
to the ones that are still running. Seems simple enough to
me.
If you want to look at a product that does this, see the
RAINcluster product at http://www.znyx.com -- you can
use it to hook a server to a Fast EtherChannel switch,
or to hook two or more servers together with trunked
lines.