THE FOLD / BOSS / THE WALL / THE BANDWIDTH DELAY PRODUCT
THE BANDWIDTH DELAY PRODUCT
how much you can be wrong about at once
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A fast link over a long distance holds an enormous amount of data in flight. If the sender’s window is smaller than what the path can hold, the link is idle no matter how fast it is.
LIT verified live. A 10,000 Mbps link at a 100 ms round trip holds 125,000,000 bytes in flight — 122,070.3 KiB. With the classic 64 KiB window the achievable throughput is 5.24 Mbps, which is 0.05% of the link. Nothing is congested, nothing is lost, and the wire is almost entirely empty because the sender has run out of permission to speak.
LIT verified live. A 10,000 Mbps link at a 100 ms round trip holds 125,000,000 bytes in flight — 122,070.3 KiB. With the classic 64 KiB window the achievable throughput is 5.24 Mbps, which is 0.05% of the link. Nothing is congested, nothing is lost, and the wire is almost entirely empty because the sender has run out of permission to speak.
2 HOW IT WAS WEAVED · AI + HUMAN
This is why TCP window scaling exists at all, and why the 64 KiB ceiling of the original header became a hard limit on long fat networks.
AVAN (AI) computed the achievable rate rather than the shortfall, because a percentage hides the shape. 5.24 Mbps on a 10 Gbps link is the same number at every distance-limited point: with a fixed window, throughput depends only on the round trip and not on the link at all.
AVAN (AI) computed the achievable rate rather than the shortfall, because a percentage hides the shape. 5.24 Mbps on a 10 Gbps link is the same number at every distance-limited point: with a fixed window, throughput depends only on the round trip and not on the link at all.
3 ONE DIMENSION
What the path holds, against what a window allows.
4 TWO DIMENSIONS · INTERACTIVE
Change the link, the distance, and the window.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: a pipe mostly empty.
AVAN’s addition (the inverse-companion): the forward reading is that a small window wastes a fast link. The inverse is that the window is not a limit on sending, it is the amount of loss you are willing to have caused before you find out. Everything in flight is unacknowledged, which means everything in flight may already be gone. Read backwards, 122,070.3 KiB of window is 122,070.3 KiB of exposure, and the number that makes the link fast is the same number that decides how much you can be wrong about at once.
LIT a 10,000 Mbps link at a 100 ms round trip holds 125,000,000 bytes in flight, which is 122,070.3 KiB, so with the classic 64 KiB window the achievable throughput is 5.24 Mbps or 0.05% of the link - nothing congested, nothing lost, and the wire almost entirely empty because the sender has run out of permission to speak
FIG This is why TCP window scaling exists at all, and why the 64 KiB ceiling of the original header became a hard limit on long fat networks. AVAN computed the achievable rate rather than the shortfall, because a percentage hides the shape. 5.24 Mbps on a 10 Gbps link is the same number at every distance-limited point: with a fixed window, throughput depends only on the round trip and not on the link at all.
FIG This is why TCP window scaling exists at all, and why the 64 KiB ceiling of the original header became a hard limit on long fat networks. AVAN computed the achievable rate rather than the shortfall, because a percentage hides the shape. 5.24 Mbps on a 10 Gbps link is the same number at every distance-limited point: with a fixed window, throughput depends only on the round trip and not on the link at all.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE WALL · David Lee Wise (ROOT0), with AVAN