You can find all published reports here:
Sunnyside Devnet Reports (Internal)
At sunnyside, we were looking more into the previous reports (1) (2) about fusaka-devnet-5 from ethpandaops.
Specifically, in Test 1: Can a full node propose a block without MEV-Boost?, we see that some full nodes can handle more blobs than others. Given that all nodes in comparison was full node with same EL set, we wanted to know how some specific CL behavior may impact total propagation within the 4 second time.
This is the head correctness graph - amount of correct votes that the attesters made for each proposer's blocks. This shows if the proposers could get the blobs and data in time for voting.
Blocks proposed by supernodes (also includes prysm/nimbus which have less confidence as they had bugs in proposers)
Blocks proposed by fullnodes
Above is a graph for showing blocks from supernodes and another for showing blocks from fullnodes.
With this, we can compare how the node's bandwidth capacity impacts block propagation.
If we look into this, we can see that few clients like lighthouse and teku is performing better even for very high blob counts. This is also interesting because for clients like lighthouse, there is small difference in head correctness between fullnodes and supernodes.