RE: RE: RE: Load Balancing Across Trunks

From: Tim (ccie2be@nyc.rr.com)
Date: Tue Sep 13 2005 - 23:18:55 GMT-3


Ralph,

In the lab, you will have complete control over both your switches but you
will not have access to any backbone routers.

So, let's say this is the topology:

BB == sw1 == sw2

Now, let's assume you have multiple links between all 3 devices and the BB
is acting like a switch and is configured to be the root switch. Since you
don't have access to the BB, if you have to configure sw1 in such a way that
you specify which vlan's are forwarded over which ports and you're not
allowed to use the cost parameter, you've got yourself a tricky situation.
But, I'll get back to that shortly.

If you need to specify which vlans will be forwarded over which links
between sw1 and sw2, that's easy. First determine on which switch your
going to config to accomplish this, then determine which parameter to use.

Conversely, if they tell you which parameter to use but not which switch,
just remember where each parameter is used and configure accordingly.

Now, let's get back to the trunks and links betw the BB and SW1. Let's also
say that you have to use the port priority parameter to implement the
requirements. Typically, the port priority parameter is only configure on
the upstream switch but in this case, the upstream switch is BB and you
don't have access to it. Now what?

At first blush, this might sound like a hopeless situation but in the lab,
unless they screwed up royally (which is highly unlikely), there must be a
way to accomplish this.

So, the first thing I would do is try to think of some way to make sw1 the
root switch. Since this is a STP function, I would enter, "span ?". Doing
that, you'll find a command that allows you to make sw1 the root bridge.

So, you do that but it doesn't work. BB remains the root bridge. Now, what?

Hmmm, this is getting tougher. How about looking in the chapter in the 3550
software config guide on Optional Spanning Tree blah, blah?

If I'm not mistaken, you'll find something regarding root guard. So, I would
try that.

Now, if that doesn't work, I'm out of ideas. And, in such a situation, I
would move on to the next task in the lab and try to get those points.

HTH, Tim

-----Original Message-----
From: Ralph [mailto:Mandela@myrealbox.com]
Sent: Tuesday, September 13, 2005 9:31 PM
To: ccie2be@nyc.rr.com
Cc: terry.francona@gmail.com; ccielab@groupstudy.com
Subject: Re: RE: RE: Load Balancing Across Trunks

Good explanation Tim. My question is based on the fact that the original
post did not specifically say how the switches are connected, (that
information is not given). In fact to really build some fault tolerance into
a switching network using spanning tree you will have to connect the
switches together into a loop so that if a port or a switch fails, traffic
for vlans can switch over to other ports or switches. In your topology if
sw2 fails what happens?

As regards to how the swithces are connected, is it okay to make such an
assumption in the lab, or just use a different way to achieve the same
result without making assumptions?

Regards
Ralph

-----Original Message-----
From: "Tim" <ccie2be@nyc.rr.com>
To: "'Ralph'" <Mandela@myrealbox.com>
Date: Tue, 13 Sep 2005 20:51:09 -0400
Subject: RE: RE: Load Balancing Across Trunks

Hey Ralph,

If you really want to completely understand this, I strongly recommend you
read the classic Cisco Press book on this topic, LAN Switching by Kennedy
and Clark.

But, I'll try to summarize the concept for you here.

STP reason for existence and the reason it was created is to create a L2
LOOP FREE path between every device in the L2 domain.

Step 1 in this process is to elect a root bridge. Once a root bridge is
elected, every other switch in the L2 domain has to find the lowest cost
path to the root bridge.

So, going back to the previous example with the 3 switches, once sw3 is
elected the root bridge, sw2 which has multiple paths to sw3, the root
bridge, has to figure out which physical link it will use to reach the root
bridge, sw3.

If I remember correctly, (a big if indeed), sw2 will first compare the costs
of each path to the root bridge. If these costs are equal, then it compares
port priority. If these are equal, the tiebreaker is interface number.
Based on this assessment, sw2 will put 1 of its ports into a forwarding
state and all the other ports into blocking state.

After sw2 is done deciding which port will forward and which will be
blocked, sw1 will do the same thing. Recall that sw1 also has multiple
paths (multiple physical links connecting it to sw2). And, in our example
topology, the only way for traffic to reach the root bridge from sw1 is via
sw2.

BTW, sw1 and sw2 know which way to go to get to the root bridge because of
bpdu's that flow downstream from the root bridge. ( If you want to know the
details of this process you definitely have to read the Lan Switching book
cause I don't remember).

So, now that sw1 knows that it has to get to the root bridge via sw2, it has
to decide which port it will put into the forwarding state and which port(s)
into the blocking state.

So, just like sw2 did, sw1 will compare cost, priority and then, if needed,
interface number.

As an aside, since in the lab, there are only 2 switches, you don't have to
worry to much about this. However, you better know how to manipulate which
port goes into forwarding and which are blocked and when to configure which
parameter on which switch.

I use this memory aid:

D U D

C P I

Where D = downstream; U = Upstream; C = cost; P = priority; and I =
interface #

So, if you have 4 physical 100mbps links between 2 switches and you want
port x to forward, you can force this to happen by either manipulating cost
on the downstream switch (make it lower on port x) or increasing the port
priority on the upstream switch.

Make sure you know this inside out, backwards and forwards, and in your
sleep.

HTH, Tim

-----Original Message-----
From: Ralph [mailto:Mandela@myrealbox.com]
Sent: Tuesday, September 13, 2005 8:00 PM
To: ccie2be@nyc.rr.com
Cc: terry.francona@gmail.com; ccielab@groupstudy.com
Subject: Re: RE: Load Balancing Across Trunks

Tim, I do agree that between sw1 and sw2, you can use port-priority, but how
exactly would you make sw2 the upstream switch; since either of them has a
chance of being the upstream switch, without using cost ?

Regards
Ralph

-----Original Message-----
From: "Tim" <ccie2be@nyc.rr.com>
To: "'Ralph'" <Mandela@myrealbox.com>, <terry.francona@gmail.com>
Date: Tue, 13 Sep 2005 18:37:29 -0400
Subject: RE: Load Balancing Across Trunks

Guys,

With all due respect, I disagree with this thinking.

I haven't tested it, so I won't comment on the effectiveness of using
port-priority on the upstream non-root switch relative to the downstream
switch.

But, I'm sure using cost on the downstream switch will work.

Actually, now that I think about it, I'm pretty sure using port priority
will also work even when neither switch is the root switch because if there
are more than one physical links between the 2 non root switches, STP still
has to determine which physical link to put in block state.

So, let's say you have this topology:

Sw1 sw2 sw3

And between each pair of switches there are multiple physical links. Lets
also assume that sw3 is elected the root switch. Then relative to the root
switch, sw2 is the upstream switch from sw1's point of view.

Now, how does STP decide which physical link between sw1 and sw2 should be
forwarding and which should be blocking?

Wouldn't it go through the exact same process that sw2 and sw3 did?

tim

-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
Ralph
Sent: Tuesday, September 13, 2005 5:54 PM
To: terry.francona@gmail.com
Cc: ccielab@groupstudy.com
Subject: Re: Load Balancing Across Trunks

Yes Anthony, if you do not have control or access to the root switch for a
vlan there is really very little you can do to load balance traffic for that
vlan. If etherchannel is not allowed, you can use the switchport trunk
allowed vlan command; But I'm not sure if this is fault-tolerant because it
is not a spanning tree feature.

Ralph.

-----Original Message-----
From: Anthony Sequeira <terry.francona@gmail.com>
To: Group Study <ccielab@groupstudy.com>
Date: Tue, 13 Sep 2005 17:28:31 -0400
Subject: Load Balancing Across Trunks

It seems to be a simple task to load balance traffic on a VLAN basis across
your trunk links if you are dealing with only two switches that you
completely control. For example, if you are forbidden from using port cost,
just make one of your two switches the root for all VLANs and then set the
port priorities apropriately on this upstream switch for each VLAN.
 But what if the root of a VLAN you need to load balance is on a third
switch out of your control? Now you can play with port-priority all you want
on your two switches but your configurations will have no effect.
 Must we be able to control the root switch election in order to properly
load balance across trunk links using port priority? I have "labbed" this up
- and it seems that we do need this level of control.
 Is there another way to control load balancing across trunk links beyond
port cost and port priority? I think not.....



This archive was generated by hypermail 2.1.4 : Sun Oct 02 2005 - 14:40:15 GMT-3