Posted by t-bast
May 30, 2025/12:51 UTC
The discussion touches upon the technical nuances of TCP packet sizes and their implications on network performance. The initial point raised concerns the assumption that TCP packets, ideally sized at 65kB, do not undergo fragmentation by intermediate routers. This assumption leads to an open-ended question about the practicality of such packet sizes, especially considering whether they often get fragmented, the overhead this might introduce, and the lack of public data on the subject. The conversation reveals a concern that if substantial fragmentation occurs, the result could be degraded performance due to both the fragmentation itself and increased bandwidth usage.
It is noted that, in practice, TCP packets are typically limited to around 1500 bytes. This limitation prompts questions regarding whether such a size is dictated by default operating system behavior or merely a recommendation. The absence of easily accessible data on current internet practices regarding TCP packet handling is acknowledged, alongside a proposal for an experimental approach to gather empirical data.
The proposed experiment involves conducting A/B testing on mainnet nodes by alternating between padding all packets to 65kB and removing the padding, while measuring latency differences. The goal of this experiment would be to obtain some insights into the overhead associated with packet padding and fragmentation, albeit with the understanding that current payment volume may not be sufficient to provide statistically meaningful results. This method is suggested as a starting point to better understand the impact of packet size decisions on network performance.
TLDR
We’ll email you summaries of the latest discussions from authoritative bitcoin sources, like bitcoin-dev, lightning-dev, and Delving Bitcoin.
We'd love to hear your feedback on this project?
Give Feedback