THE FOLD / BOSS / THE WALL / THE WRITE AMPLIFICATION
THE WRITE AMPLIFICATION
each layer answered honestly a question it was asked wrong
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Change one byte. The filesystem writes a page, the journal writes it twice, the SSD programs a page of its own, and the garbage collector eventually copies a whole erase block to reclaim the space.
LIT verified live, on a stated model. 1 logical byte becomes 4,096 after the filesystem page, 8,192 after the journal, 16,384 after the SSD page, and 28,672 once garbage collection copies a 75%-live erase block. That is 28,672×. Every layer is behaving correctly and none of them is aware of the others.
LIT verified live, on a stated model. 1 logical byte becomes 4,096 after the filesystem page, 8,192 after the journal, 16,384 after the SSD page, and 28,672 once garbage collection copies a 75%-live erase block. That is 28,672×. Every layer is behaving correctly and none of them is aware of the others.
2 HOW IT WAS WEAVED · AI + HUMAN
Write amplification is why SSD endurance is quoted in drive-writes-per-day and why database people care enormously about page size.
AVAN (AI) is stating plainly that these are declared parameters — 4K page, 2× journal, 16K SSD page, 4 MiB erase block, 75% live — and not a measurement of any real drive, which a browser cannot reach. What is demonstrated is that the factors multiply, which is a property of the stack rather than of any particular hardware. Change the parameters on the page and watch the product move.
AVAN (AI) is stating plainly that these are declared parameters — 4K page, 2× journal, 16K SSD page, 4 MiB erase block, 75% live — and not a measurement of any real drive, which a browser cannot reach. What is demonstrated is that the factors multiply, which is a property of the stack rather than of any particular hardware. Change the parameters on the page and watch the product move.
3 ONE DIMENSION
One byte, through four layers.
4 TWO DIMENSIONS · INTERACTIVE
Change a layer and watch the product.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: one byte, widening.
AVAN’s addition (the inverse-companion): the forward reading is that the stack wastes 28,672 bytes to store one. The inverse is that nobody wasted anything — each layer was told a lie by the one above and answered it honestly. The filesystem was asked to store a byte and can only address pages. The SSD was handed a page and can only program its own. Read backwards, the amplification is not inefficiency but the cost of every layer being allowed to define its own smallest unit, which is the same freedom that let any of them be built separately at all.
LIT on a stated model 1 logical byte becomes 4,096 after the filesystem page, 8,192 after the journal, 16,384 after the SSD page and 28,672 once garbage collection copies a 75%-live 4 MiB erase block - a total of 28,672x - with every layer behaving correctly and none of them aware of the others
FIG Write amplification is why SSD endurance is quoted in drive-writes-per-day and why database people care about page size. AVAN is stating plainly that these are declared parameters - 4K page, 2x journal, 16K SSD page, 4 MiB erase block, 75% live - and NOT a measurement of any real drive, which a browser cannot reach. What is demonstrated is that the factors multiply, which is a property of the stack rather than of any hardware.
FIG Write amplification is why SSD endurance is quoted in drive-writes-per-day and why database people care about page size. AVAN is stating plainly that these are declared parameters - 4K page, 2x journal, 16K SSD page, 4 MiB erase block, 75% live - and NOT a measurement of any real drive, which a browser cannot reach. What is demonstrated is that the factors multiply, which is a property of the stack rather than of any hardware.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE WALL · David Lee Wise (ROOT0), with AVAN