Increase g_in to 500ms

[karknu]

Increase g_in to 500ms. This increases the pipelining performed by blockfetch which means that we better utilize the network when talking to far away nodes.

Goes great with IntersectMBO/cardano-node#708

[dcoutts]

Karl reports the following numbers for blocks synced in 5 minutes with the two settings

In this test the nearby node is in the EU and the far away node is in Asia. The test node of course is in the EU.

Test relay ping times (measured using tcpping):

• Nearby: 26ms
• Far away: 240ms

Results

Blocks synced in 5 minutes:

Current value: g_in 100ms

• Nearby node: 156,320
• Far away node: 42,361

Proposed value: g_in: 500ms

• Nearby node: 146,069
• Far away node: 119,002

Comments

So this shows that the bigger g_in brings the sync speed for the far away node up very significantly. Almost as important, the sync speed for the far node is now 80% of that of the near node, whereas before it was ~27% the speed. So this means we should expect much less variance in sync speed depending on which relay you get from the DNS pool. This is good for the time until we get the performance-based relay selection implemented and integrated.

Finally, it's another demonstration that our network protocols -- that we designed to be tolerant of network latency -- really are so. The latency difference between the near and far nodes is a factor of 10 here.

[karknu]

bors r+

[iohk-bors]

Build succeeded

• ci/hydra:Cardano:ouroboros-network:required
• buildkite/ouroboros-network