Load balancing on PVCs

[kapishmohole.cisco]

Hello,

I have three PVCs for same destination, all are of equal bandwidth (sub interfaces). And all are directly connected. Three static routes are used for routing packets to the network connected to the next router, all three routes have default AD.

Now I want equal load balancing on these three PVCs.

How can I achieve it?

Using ‘ip route-cache’, is it possible? But I guess this command doesn’t work for sub interfaces. Is there any other way, by changing parameters like fast switching/routing cache etc?

We are not using any routing protocol, and not supposed to use it.

[marikakis]

There are two directions for load-balancing.

From (lets say) R1 to R2, and from R2 to R1.

You need to add static routes on both routers

for load-balancing to work in both directions.

Static routes on R1 should cover all networks that can be reached via R2,

and static routes on R2 should cover all networks that can be reached via R1.

For example, lets suppose R1 is a default gateway for R2.

Further suppose that R2 can reach only the extra network 10.0.0.0/8

On R2 you can add 3 static default routes of equal AD towards getaway R1.

On R1 you can add 3 static routes to network 10.0.0.0/8 of equal AD via R2.

The type of load-balancing will depend on the switching mechanism used.

Usually, it doesn't matter if you are working with subinterfaces.

I can only recall a case where CEF doesn't work

when switching between dot1q subinterfaces.

Considered different feature from pure CEF.

I have seen no problem with ATM subinterfaces.

However, switching mechanism availability depends on the platform.

Probably fast-switching with per-destination load-balancing

is already enabled in your router.

To verify the switching mechanism used on your router's interfaces

use the "show ip interface" command and look for the word "switching".

I would try CEF.

Since you are not running a routing protocol,

I suppose CEF will not consume much memory.

Try the default per-destination load-balancing

(actually is per source/destination).

To enable CEF, use the command "ip cef" in global configuration mode:

If you have disabled CEF operation on an interface and want to reenable it,

you can do so by using the "ip route-cache cef " command from interface configuration mode.

If you need finer distribution of traffic in low bandwidth links,

you can enable per-packet load-balancing

("ip load-sharing per-packet" from interface configuration mode on all

relevant subinterfaces),

at the cost of probably some increase in CPU usage and

chances of packets arriving re-ordered at the destination.

M.

[kapishmohole.cisco]

Hello,

I have one more problem in load balancing theory. I have 3745 router, so fast packet and CEF both are running by default on it.

Which one is actually handling the load balancing, CEF or fast packet?

In ‘show ip int’ command I see both fast packet and CEF both are running. Is it like one overrides another or both techniques do the load balancing?

[rwiesmann]

Hi

Basically there are two types of Load Balancing. Per Packet and Per Destination.

Per Packet is Packet by Packet and uses more resources.

Per Destination send alls Packets for a given destination over the same link.

Per Packet Load Balancing is inappropriate for certain type of data traffic that depend on arrival of packets in sequenze like voice traffic over IP.

If you have three static routes for a destination with the same AD laod balancing should already work.

Check with "sh ip route". All three static route should be in the routing table.

With the ip route-chache you can enable per packet or per destination load balancing. Per destination ist enabled per default.

Hope that helps

Roger

[Kevin Dorrell]

Here is a document about choosing a switching algorithm for your router, and what the effects of it are:

http://www.cisco.com/en/US/tech/tk827/tk831/technologies_white_paper09186a00800a62d9.shtml

Kevin Dorrell

Luxembourg

[kapishmohole.cisco]

Hello,

Thanks to you all for this great information.

I will be using CEF for this purpose.

But I have one more question – If suppose any two out of these three PVCs goes down, what will happen with routing?

I have seen one scenario of load balancing, when any connection goes down, all users going through that PVC gets disconnected, they can’t reach the destination, even though there are other ways for destination. Mostly it happens with per destination basis load sharing where packets to one destination go through one link only.

Is it possible to maintain connectivity for all users even if any link goes down? I guess per packet basis load sharing will work.

What is best, per packet OR per destination basis? What are additional things I should do or take care of?

[marikakis]

Make sure your PVCs are managed by OAM.

See the following link for details:

http://www.cisco.com/en/US/tech/tk39/tk48/technologies_configuration_example09186a0080094570.shtml

The choice depends on the bandwidth of the PVCs.

For low bandwidth links, per-packet is usually better.

But per-packet will probably increase the CPU usage in your router

compared to the per-destination flavor.

For high bandwidth links, it is not usually worth

the effort to try to finely distribute traffic

with the per-packet flavor.

You can try per-destination,

carefully monitor how it goes throught the hours,

and use per-packet if you really need it.

M.

[kapishmohole.cisco]

Hello,

Its great thing you told, OAM.

I have faced similar incident; PVC is up but no data flowing on it. After resetting UFM card of IGX, PVC started to work.

I have 3745 router and IGX, so I am directly connecting router to IGX using smart serial interface. And many sub interfaces/PVCs on that interface.

I don’t have ATM interface on router, and OAM will not work for me. Isn’t it? Its ATM stuff.

I didnt find any command for OAM on serial interface.

But to avoid effect of PVC malfunctions and drop of traffic, can I use OAM on smart serial interface, or is there any other method for this purpose?

[marikakis]

You are right, OAM is for ATM.

I somehow thought you had ATM PVCs, although you never said that ;-)

I am sorry.

The only problem I have come across (I year ago),

was with PVCs where the one endpoint was on an FR interface

and the other was on an ATM interface (i.e. ATM/FR PVCs).

I remember that the problem was caused by something not working well on the IGX/BPX side,

but I was on the router part and do not exactly know what was wrong with the switch.

Maybe now there exists some solution.

I will look around and if I find something I will inform you.

M.

[kapishmohole.cisco]

OK,

Thanx for the info...

Kapish

[ashishrohilla]

The link which u provided for oam does not describe that it can be configured when frame-relay is used. I am using IGX as frame-relay switch, can we use oam with IGX frame-relay PVCs.

[marikakis]

I suppose you fellows are both interested in Frame Relay only.

There is no equivalent of OAM for Frame Relay.

However, there is a cisco proprietary feature you might try.

It is called Frame Relay end-to-end keepalive.

See the following link for details:

http://www.cisco.com/en/US/products/sw/iosswrel/ps1830/products_feature_guide09186a0080087a58.html#5038

The 3700 series is not listed under the supported platforms,

but the document is for 12.0T and probably old.

You can check if the relevant commands can be entered in your router.

M.