Per packet vs per flow routing

But what will happen on per flow if one of the T1's fills up with traffic, is there a way it would know to switch to other T1?

Should I c onsider a ppp multilink?

No. No routing protocol exists that takes load into account. Yes, EIGRP *can* use it as a metric, but it's zeroed out by default. Why do you suppose that is? The ability to detect congested links and react to it intelligently is quite challenging from a routing perspective.

That's one way to solve it if you have CPU to spare.

Is there another way to solve it?

Tx

Unless you have a single source/destination pair that needs more than a T1 of bandwidth than flow based load balancing will be fine. If this won't work, than you can turn on per packet load balancing but it will cost you CPU cycles. If you have a low end router be careful. The reason load balancing defaults to per flow is because it is less CPU intensive and it works just fine 99% of the time.

Scott

Unfortunately, it is a 3825 router and I have a voice and video requirement. This eliminates per packet load balance. The endpoints are vpn devices, so it is all single source/dest pair and I need at least 2Mbps of the bandwidth. The only option then is ppp multilink. Correct?

Thanks for your help

In article , thrill5 wrote: :Unless you have a single source/destination pair that needs more than a T1 :of bandwidth than flow based load balancing will be fine. If this won't :work, than you can turn on per packet load balancing but it will cost you :CPU cycles. If you have a low end router be careful. The reason load :balancing defaults to per flow is because it is less CPU intensive and it :works just fine 99% of the time.

Unless, that is, you are using CeF -- modern CeF has little to no extra overhead to go per-packet. But if you are using flow caching instead of CeF, then per-packet balancing requires dropping down to process switching.

CEF is a smarter switching method common on all new gear. It uses source/destination IP among other things to create the flow.