The obvious advantage to Ethernet for ISPs is the cheaper interface cost (although not always strictly true, especially when looking at Juniper PICs) and the increased bandwidth. The flip side is that it's harder to get *fast* redundancy (no APS) without SDH, or using extremely RSTP which introduces another set of headaches.
It is actually possible to make Ethernet just as inefficient as a transport protocol as ATM
Fraktul: Hmm sub 25ms for MPLS itself, STP is slow though yes.MPLS Fast Re-Route is generally more in the 40-60msec region, rather than 25ms, in order to match SDH with ~50ms.
Also Jumbo frame support is nice for WANs.
Fraktul:Certainly getting off topic but an interesting chat anyway... I do agree w.r.t stigma. It also amuses me to no end that MPLS, Ethernet, and IP are being used to try and re-implement what ATM has done since day 1: Support multiple class of service on a single interface, and support it well. I think ATM's true power is beyond CBR services and actually in the other grades of service, and in particular being able to support discard eligibility for traffic. This is a very powerful function, particularly in designing residential broadband services.
Sure but thats true of any data layer. You pick and implement whats appropriate based on what your transporting, amongst the myriad of other requirements and trade offs. There is quite a bit of ATM stigma is seems, while some of it is deserved it does has its place. I mean for CBR services for instance its arguably technically superior
Anyhow I digress, getting off topic!