THE FOLD / BOSS / THE CHOKE POINT / THE CONVOY EFFECT
THE CONVOY EFFECT
the total wait is fixed; order decides whose it is
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
First-come-first-served is the fairest-sounding rule there is, and it is the one that makes almost everybody wait longest. A few long jobs at the front, and the whole short queue sits behind them.
LIT verified live. 200 jobs, 12 long and 188 short. FIFO gives a mean wait of 1,540. Shortest-job-first gives 409 — 3.77× shorter. The total work is identical: both finish at 3,689. Nothing was made faster and nothing was dropped; the same jobs ran on the same machine for the same total time, and the only thing that changed was the order.
LIT verified live. 200 jobs, 12 long and 188 short. FIFO gives a mean wait of 1,540. Shortest-job-first gives 409 — 3.77× shorter. The total work is identical: both finish at 3,689. Nothing was made faster and nothing was dropped; the same jobs ran on the same machine for the same total time, and the only thing that changed was the order.
2 HOW IT WAS WEAVED · AI + HUMAN
The convoy effect is the standard argument for SJF, and the standard argument against it is that it needs to know the job lengths — which is what THE MULTILEVEL FEEDBACK gets around.
AVAN (AI) published the makespan next to the mean wait because that is the column that shows nothing was stolen. 3,689 either way. A scheduler cannot create throughput; it can only decide who does the waiting, and FIFO decides that 188 short jobs should wait for 12 long ones.
AVAN (AI) published the makespan next to the mean wait because that is the column that shows nothing was stolen. 3,689 either way. A scheduler cannot create throughput; it can only decide who does the waiting, and FIFO decides that 188 short jobs should wait for 12 long ones.
3 ONE DIMENSION
The same jobs, two orders.
4 TWO DIMENSIONS · INTERACTIVE
Reorder the queue and watch the wait, not the finish.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: a queue behind one long thing.
AVAN’s addition (the inverse-companion): the forward reading is that SJF is better than FIFO. The inverse is that it is better for the mean and worse for the long job, and the mean has no opinion about that. Every one of those 188 short jobs gained, and the 12 long ones paid for all of it. Read backwards, “3.77× improvement” is a statement about a population, and there is no arrangement of a queue that improves it for everybody — the total wait is fixed by the job lengths, and a scheduler only ever decides whose it is.
LIT 200 jobs - 12 long and 188 short - give a FIFO mean wait of 1,540 against 409 for shortest-job-first, which is 3.77 times shorter, while the total work is identical at a makespan of 3,689 either way, so nothing was made faster and nothing was dropped and the only thing that changed was the order
FIG The convoy effect is the standard argument for SJF, and the standard argument against it is that it needs to know the job lengths - which is what THE MULTILEVEL FEEDBACK gets around. AVAN published the makespan next to the mean wait because that is the column showing nothing was stolen: 3,689 either way. A scheduler cannot create throughput, it can only decide who does the waiting, and FIFO decides that 188 short jobs should wait for 12 long ones.
FIG The convoy effect is the standard argument for SJF, and the standard argument against it is that it needs to know the job lengths - which is what THE MULTILEVEL FEEDBACK gets around. AVAN published the makespan next to the mean wait because that is the column showing nothing was stolen: 3,689 either way. A scheduler cannot create throughput, it can only decide who does the waiting, and FIFO decides that 188 short jobs should wait for 12 long ones.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE CHOKE POINT · David Lee Wise (ROOT0), with AVAN