Output Of Local Example Different Than README + Help Understanding Values about p2plab

[hinshun]

@bonedaddy Btw, I added an additional section on benchmarking the quic libp2p transport: https://github.com/Netflix/p2plab#live-updating-the-cluster. More to come later.

[hinshun]

This is just a rendering feature from the table writer: https://github.com/olekukonko/tablewriter
Rows that have the same value have the cells merged. i.e. in the README's benchmark the neighbor nodes both received 0 blocks and 0 duplicate blocks, whereas in your run the neighbor nodes received a few duplicate blocks.

[bonedaddy]

Ah okay, makes sense. Thanks.

[bonedaddy]

@hinshun Thanks! Another question I have is determining what kind of results are better than others. For example in the bitswap tests do we want BLOCKSRECV to be equal across all nodes, and as low as possible? For the bandwidth tests do we want the fastest mb/s as possible?

Thanks.

[hinshun]

It depends on your scenario and what you care about. In this specific example, you probably want:

• Minimize total time for all (not 'neighbor') nodes to receive the DAG
• Minimize total DUPBLOCKS, aka minimize BLOCKSRECV
• Minimize BLOCKSRECV in neighbor nodes (they have the full DAG so they shouldn't need any blocks)
• Balance the BLOCKSSENT in neighbor nodes (so load is distributed)

That said, the bitswap implementation should perform well generally. So if a change in bitswap improves one scenario but greatly slows down in another, then it may not be acceptable. In practice, you probably want to write a suite of scenarios, assign a weight to each scenario, and then compute a final score.

[bonedaddy]

Makes sense, thanks.