Straight from Cisco’s Campus Network for High Availability Design Guide, L3 is superior because:
– Faster convergence around a link or node failure.
– Increased scalability because neighbor relationships and meshing are
– More efficient bandwidth utilization.
From Cisco: Campus Network for High Availability Design Guide - Core Layer
These are all great reasons, but you won’t stand to benefit from any of them, as your setup doesn’t allow for it. If a packet arrives at your distribution node, it’s only going to have one node that it can go to next; your core which is using IRF. There isn’t a routing protocol that is going to be intelligent enough to determine that you have another (perhaps lower speed) connection combining your 2 cores. That additional hop is transparent and would require some manual intervention (somehow).
Personally, I don’t think it’s a good idea for you to virtualize your cores. Path determination is impossible, independent device operation is severed, and you completely eliminate the possibility of intelligent routing. All of those are critical to a well oiled network.
Though, you may have additional business requirements that dictate this design but aren't outlined above.
In the Cisco world, it's always been highly recommended to have layer
3 between distribution and core using a routing protocol.
This isn’t just the Cisco world, this is the Networking world overall. Network engineers work to eliminate STP in as many cases as possible because they know of the limitations it drags along with it. Routing to the access layer is recommended nowadays:
Consider EIGRP/Routing in the access layer.
A routing protocol like EIGRP, when properly tuned, can achieve
better convergence results than designs that rely on STP to resolve
convergence events. A routing protocol can even achieve better
convergence results than the time-tested L2/L3 boundary hierarchical
design. However, some additional complexity (uplink IP addressing and
subnetting) and loss of flexibility are associated with this design
alternative. Additionally, this option is not as widely deployed in
the field as the L2/L3 distribution layer boundary model.
From Cisco: Campus Network for High Availability Design Guide
- Access Layer Tuning
There used to be an idiom: “Switch where you can, Route where you must”. In recent times, that has been completely flipped on it’s head:
Route where you can, Switch where you must
Tons of enterprise switches support some level of routing now. We can’t stand behind ‘my device doesn’t support that’ because that’s not the case, anymore.
Is the practice of implementing a 'layer 3 distribution to core'
applicable today in the HP world when it's possible to have a highly
redundant layer 2 to the core without STP, using IRF?
Yes, it absolutely is. HP’s Intelligent Redundant Framework is just another system virtualization technology like Cisco’s StackWise and Juniper’s Virtual Chassis, just painted a different color and tagged with ‘the most disruptive technology in ages’. Every vendor will tout how much better theirs is than the competitor, but it’s not a game changer or anything to write home about.