We attended the Ethereum 2.0 networking call. Kudos to Daniel and Jason to be up at the insane hour of 5am.
Here is the content they shared during the meeting:
High Level to Present
- The previous preliminary results resulted in poor data because of an implementation bug, which had been rectified and the tests were rerun.
- There was no resource throttling that was observed across all tests
- Overall, libp2p’s gossipsub seems to perform as intended
- The 400ms latency test yielded an average msg_delivery time of 1.2 seconds (with the BA degree of 2).
- The maximum time found across tests was 4.573011459 seconds. This is still under the 6 second block time, which is good
- The last_delivery_hop seemed to be dependent on the degree of connectivity. But anything greater than a degree of 2 plateaued out and seemed to have little to no affect. Further testing needs to be done to verify this.
- The delivery time graphs showed interesting behaviors where they would look similar to a damped sinusoidal wave. The causes of the “lobes” cannot be pinpointed exactly, however those lobes were not present in tests where the message rate was low (20msg/sec).
- one possible reason for the lobes could be that the initial spike were the messages that were put first into the queue. then the second lobe forms from the second group of messages that were propagated on the next heartbeat. Then another group on the next heartbeat, until the queue had been emptied.
More information on the call detailed here:
As always, Ben Edgington did a good job at capturing notes here: https://hackmd.io/@benjaminion/SJ3W0qwAH