What is a realistic measure of Delta_hdr

[ch1bo]

The Leios stage length $L$ is a protocol parameter, but constrained by $L >= \Delta_{hdr}$ by the upper bound delay of transmitting a header across the network $\Delta_{hdr}$.

What is a realistic value of $\Delta_{hdr}$?

In the paper and various protocol sketches it is defined that the ${hdr}$ in question is the IB header, EBs, or votes (the latter two are header-only). So I assume the transmission delay depends on how big those messages are assumed to be vs. the considered network size (characterized by capacity and diameter?).

[nfrisby]

Today's network has working estimate of $\Delta$ (not just $\Delta_{hdr}$). It seems plausible to me that that's an upper bound for $\Delta_{hdr}$, based on the idea that (full) Praos blocks are much bigger than any Leios header. cc @coot

[ch1bo]

Is the current $\Delta$ the 3 or 5 seconds?

[coot]

It's 5 seconds; 99.5% of blocks arrive in <1s, 99.8 in <3s,

[ch1bo]

This was discussed in today's AMA and what I understood:

• The praos $\Delta Q$ model could be a good indication for the upper bound in use of praos headers
• Praos block headers are about 1100 bytes in size; we are careful to fit them in one MTU
• The storage and the CIP draft do give an indication that Votes would be smaller and EBs (or EB headers) likely also fit that requirement.
• Praos block headers take between 3 seconds (to reach 99%+ nodes) to 1 second (to reach ~90%+ of stake weighted nodes)

So at least the order of magnitude we can count with is around 1-2 seconds.

[coot]

Praos block headers take between 3 seconds (to reach 99%+ nodes) to 1 second (to reach ~90%+ of stake weighted nodes)

That's true for blocks; even more so for headers, but we don't measure that.