Just a bit of a discussion on the previous post.
Why would you use this, on a slightly more serious note than before – in all honesty I think you probably wouldn’t. But the point is it gives you the same kind of per-flow control that a lot of people seem to be looking forward to getting from SDN. It seems silly to me, but hey, if you want it, you can get it now.
The other point, is that by using MPLS-TE you only need to define the flow at its ingress point into the MPLS network. Using straight up vanilla PBR and you would have to configure all the links in the path.
Potentially in a very large scale data centre, you could – but I don’t see it happening. A more practical use case would be a large-scale MPLS WAN. If you have a large non-hierarchical WAN. Possibly you have a large number of links for redundancy but the layout does not use a nice campus-style 3 layer design due to geographic or optical fibre constraints. Due to the layout you will end up with some links completely underused – you could potentially tweak your per-link IGP metrics – but this might not even work depending on the topology.
Potentially you might have specific services that are bandwidth intensive – say, CCTV with offsite storage? Remote office data centre backups? Services that can be easily defined and may not be highly user-interactive so some added latency won’t be an issue.
But realistically, if you are that constrained for bandwidth, I think you will have other issues. Also remember if you are using all your redundant links – then they are no longer redundant links. It should be unnecessary for me to say that, but I often hear people complaining about spanning-tree blocked links, standby firewalls or backup Internet links complaining that they aren’t getting any use out of paid for resources. Well you are, it’s there if the primary breaks. If you use both (more than 50% anyway) and one breaks – then performance will degrade – which is not so good.