THE FOLD / CO-OP / THE SYNC / THE CAUSAL CUT
THE CAUSAL CUT
the snapshot manufactures a present rather than finding one
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A snapshot of a distributed system is a line drawn across every timeline at once. Most such lines are nonsense: they catch a message arriving that has not yet been sent.
LIT verified live. Two processes of 4 events each and 2 messages crossing between them give 25 possible cuts — exactly (4+1)×(4+1), enumerated rather than sampled. 20 are consistent and 5 are not: 80.0%. The 5 impossible ones are not unlikely or rare, they are states the system can never have been in, and a naive snapshot will happily record one.
LIT verified live. Two processes of 4 events each and 2 messages crossing between them give 25 possible cuts — exactly (4+1)×(4+1), enumerated rather than sampled. 20 are consistent and 5 are not: 80.0%. The 5 impossible ones are not unlikely or rare, they are states the system can never have been in, and a naive snapshot will happily record one.
2 HOW IT WAS WEAVED · AI + HUMAN
Chandy and Lamport’s 1985 algorithm exists to take a cut that is consistent by construction, without stopping the system.
AVAN (AI) enumerated the whole cut lattice rather than arguing about it, because the count is small enough to be exhaustive and exhaustive is a different kind of claim. The useful shape is that inconsistency is a corner of the space: cuts go wrong specifically where one process has advanced past a receive that the other has not yet sent.
AVAN (AI) enumerated the whole cut lattice rather than arguing about it, because the count is small enough to be exhaustive and exhaustive is a different kind of claim. The useful shape is that inconsistency is a corner of the space: cuts go wrong specifically where one process has advanced past a receive that the other has not yet sent.
3 ONE DIMENSION
Every cut in the lattice, marked consistent or impossible.
4 TWO DIMENSIONS · INTERACTIVE
Slide the cut and watch it become impossible.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: a lattice of possible presents.
AVAN’s addition (the inverse-companion): the forward reading is that a consistent cut is a valid snapshot of the system. The inverse is that there are twenty of them and no way to prefer one. Each consistent cut is an equally real account of “now”, and the system was in all of them and none of them; the snapshot you take is a state that may never have existed at any single instant, only one that could have. Read backwards, a distributed system does not have a present that a snapshot discovers — the snapshot manufactures one, and consistency only means it manufactured a plausible one.
LIT two processes of 4 events each with 2 messages crossing give 25 possible cuts - exactly 5 times 5, enumerated rather than sampled - of which 20 are consistent and 5 are not, and those 5 are not unlikely or rare but states the system can never have been in, which a naive snapshot will happily record
FIG Chandy and Lamport's 1985 algorithm exists to take a cut that is consistent by construction, without stopping the system. AVAN enumerated the whole cut lattice rather than arguing about it, because the count is small enough to be exhaustive and exhaustive is a different kind of claim. The useful shape is that inconsistency is a corner of the space: cuts go wrong specifically where one process has advanced past a receive the other has not yet sent.
FIG Chandy and Lamport's 1985 algorithm exists to take a cut that is consistent by construction, without stopping the system. AVAN enumerated the whole cut lattice rather than arguing about it, because the count is small enough to be exhaustive and exhaustive is a different kind of claim. The useful shape is that inconsistency is a corner of the space: cuts go wrong specifically where one process has advanced past a receive the other has not yet sent.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE SYNC · David Lee Wise (ROOT0), with AVAN