This is a summary of an experiment that was run in Thunderdome. Thunderdome is our tool for comparing performance of different versions, configurations or deployments of IPFS gateways. More information here.

This experiment was designed to compare performance of various values for the bitswap provider search delay setting to help identify an appropriate setting. Five variants were run.

Key Result: No discernible effect was found from any value of provider delay setting

Performance problems were found in the restricted datastore variant but this was caused by Kubo entering a state with prolonged elevated numbers of peers, severely impacting gateway perfomance.

Details

Five variants of the experiment were run:

• base: two instances of each provider delay 100, 400, 700, 1000ms
• acceldht: as base but Experimental.AcceleratedDHTClient=false to determine if the discovery client has any influence
• datastore: as base but Datastore.StorageMax=1GB to force more bitswap requests
• peerset: as base but Swarm.ConnMgr.HighWater=600 and Swarm.ConnMgr.LowWater=300 to examine settings closer to desktop use
• long: two instances of each provider delay 10s, 20s, 30s to simulate effectively disabling provider search

All instances are sent the same requests at a rate of 20rps

Explore these results further on Grafana:

• base: Report / Timeline
• acceldht: Report / Timeline
• datastore: Report / Timeline
• peerset: Report / Timeline
• long: Report / Timeline

There was no discernible difference in P99 Time to First Byte in the baseline experiment. These values are typical for the Kubo 0.16 pre-release