cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
1132
Views
6
Helpful
21
Replies

Multilink loadbalancing

royalblues
Level 10
Level 10

I would like to know how the multilink will behave if i have 3 links bundled together each of diffrent bandwidth.

Eg. i ahve 2 links of 1MB each and the 3rd link is a 512K link

I have been using multilink without any problems with same bandwidth links but it doesn't seem to work properly in the above scenario.

i would like to know how the links will be load balanced in the above scenario.

Is this methos recommended?

thanks

Narayan

21 Replies 21

Let me see what TAC has to say about this although iam very skeptical on their response

Narayan

narayan,

Once you have the response from the tac, please post it here.

thanks

arav

Hi, this in an interesting question so thought I'd quickly Lab it up. I configured Multilink ppp over frame-relay. Without any changes to BW per-packet loadbalancing occurs and the multilink interface has BW of 200000 (correct BW for 2 virtual-templates). When I change the virtual templates to 128k and 512k the aggregated multilink interface drops to 256k.

So, to me it looks like we are still doing per-packet load balancing (as suggested in an earlier post) and the router is making the aggregated bandwidth to be the lowest cost interface multiplied by number of interfaces in the group.

interface Virtual-Template1

bandwidth 128

no ip address

ppp multilink

ppp multilink group 1

!

interface Virtual-Template2

bandwidth 512

no ip address

ppp multilink

ppp multilink group 1

r3#show int multilink 1

Multilink1 is up, line protocol is up

Hardware is multilink group interface

Internet address is 1.1.1.3/24

MTU 1500 bytes, BW 256 Kbit, DLY 100000 usec,

Mark,

I know that manipulating the virtual template bandwidth affects the bandwidth on the multilink. I have experienced the same earlier. But still how did you manage to restrict the bandwidth on one interface?

I think manually changing the bandwidth on the interface will affect only the routing protocols.

But again what we want to know is how the loads get distributed when the true physical bandwidth of the link is different

Can you try to restrict the bandwidth to 56K by policing the traffic on one interface and try pumping traffic and see what happens

I will surely post the TAC response

Narayan

Hi all,

Below is the response from TAC

Multilink PPP, standardized in Request for Comments (RFC) 1990, is similar to load balancing techniques in that it sends packets across the individual links in a round robin fashion. However, Multilink PPP adds three significant capabilities.

Firstly, because Multilink PPP works at the link layer, it makes an Nx ISDN bundle appear as one logical link to the upper layer protocols in the router. Thus, only one network address needs to be configured for the entire Nx ISDN bundle.

Secondly, Multilink PPP keeps track of packet sequencing and buffers packets that arrive early. With this ability, Multilink PPP preserves packet order across the entire Nx ISDN bundle.

Thirdly, packet fragmentation can be enabled to split large data packets into smaller packet fragments that are individually transmitted across the links. In many circumstances, fragmentation can increase the efficiency of the N x ISDN link, and is often required for links less then 768Kbps carrying VoIP.

Multilink does not round-robin packets. It give data to the links on demand.

Several small frames may land on one link to balance a large frame on another link. A fast link draws more data than a slow link. Over time, the amount of data sent on each link is proportional to its bandwidth.

MLPPP do make a point of spreading frames across all links as much as possible, and keeping all the links equally utilized. Hence if each link takes about the same amount of time to transmit the frames given to it, the results will approximate round-robin distribution.

Narayan

Hello Narayan,

Thanks for providing such a helpful information. I have a query regarding Fragmentation that what should be the MTU size of a packet for fragmentation to be applicable to that packet in MLPPP implementation.

Regards,

Mujeeb

One must be careful in mentioning theoretical support of unequal cost load balancing in protocols like EIGRP, because nobody is known to use it.

Getting Started

Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community:

Innovations in Cisco Full Stack Observability - A new webinar from Cisco