Stats on compact block reconstructions

The recent communication highlights several critical aspects of network performance and reconstruction techniques in distributed systems. The discussion opens with an acknowledgment of the efforts made to combine prefill data across multiple HB peers, aiming at enhancing the reconstruction process. This approach is particularly relevant when considering the geographical dispersion of nodes, as suggested through the proposal of conducting an experiment with one node located significantly far from others, such as in China. The primary goal here is to evaluate the impact of packet loss on reconstruction times, especially in scenarios without prefilling.

Further inquiry focuses on a detailed examination of a scatter plot that illustrates the relationship between missing bytes and reconstruction time for nodes whose peers do not engage in prefilling. Notably, instances where more than 1 MiB of data is missing yet reconstruction times are approximately 100ms raise questions about the underlying factors enabling such efficiency. Speculation arises regarding whether peers with larger window sizes might facilitate faster data transmission, thus compensating for substantial data loss and maintaining low reconstruction times despite seemingly adverse conditions.

Additionally, there's curiosity about the potential need to adjust latency formulas to account for the cost associated with transmitting data necessary for reconstruction in failure scenarios. This suggestion stems from a desire to refine these formulas to better reflect practical realities and enhance their utility in predicting and understanding system behavior under various conditions. The conversation reflects a deep engagement with the technical nuances of network performance, emphasizing the importance of empirical testing and theoretical refinement in advancing distributed systems technology.