@ValarDragon:

• If we are approaching a block (e.g., < 25 away), and our current peer we've asked is slow, try asking from more peers, as this will become the bottleneck
• Somehow track which peers are better at serving blocks to get more from them. (We see direct peering to one peer causing significant gains / removing the bandwidth bottleneck at current state machine processing times on Osmosis. We are aiming to get this processing time 5x improved)
• Tighten the timeouts / see if we are doing more timeouts than we need. @cason: "When we receive a NoBlockResponse from a peer, we don't immediately request the same block to another peer. We instead wait for the timeout (to be double checked)."
• Metrics could be better