WEBVTT

1
00:00:00.000 --> 00:00:07.500
This overview explains how preparation and per-use costs
determine a break-even point under a stated reuse model.

2
00:00:07.500 --> 00:00:15.565
A resident artifact can repay setup over many uses, while
independently cold requests repeatedly incur that cost.

3
00:00:15.565 --> 00:00:15.567
A resident artifact can repay setup over many uses, while
independently cold requests repeatedly incur that cost.

4
00:00:15.867 --> 00:00:20.500
The candidate spends more on preparation but less on every use.

5
00:00:20.500 --> 00:00:25.200
Let n count comparable uses of the same reusable object.

6
00:00:25.200 --> 00:00:29.533
We assume the object remains valid and that these abstract costs include the

7
00:00:29.533 --> 00:00:32.667
same work on both sides.

8
00:00:32.667 --> 00:00:35.967
For one use, totals are nine and five.

9
00:00:35.967 --> 00:00:38.900
For two, they are ten and eight.

10
00:00:38.900 --> 00:00:41.567
For three, both are eleven.

11
00:00:41.567 --> 00:00:47.633
The preparation penalty is gradually repaid by saving two units per use, but at three uses

12
00:00:47.633 --> 00:00:51.533
the candidate has only reached equality.

13
00:00:51.533 --> 00:00:57.367
A strict win requires eight plus n to be less than two plus three n.

14
00:00:57.367 --> 00:01:02.300
Subtracting gives six less than two n, so n must exceed three.

15
00:01:02.300 --> 00:01:06.033
The first integer count with a strict win is four.

16
00:01:06.033 --> 00:01:12.367
State whether a gate allows equality or requires a strict improvement.

17
00:01:12.367 --> 00:01:12.379
Compare complete cost functions under an explicit reuse model.

18
00:01:12.379 --> 00:01:17.367
Compare complete cost functions under an explicit reuse model.

19
00:01:17.367 --> 00:01:24.533
A setup cost is paid once only if the artifact really
remains valid and resident across those uses.

20
00:01:24.533 --> 00:01:30.233
In a toy example, M costs 8 milliseconds to prepare and 1 per use;

21
00:01:30.233 --> 00:01:34.133
C costs 2 to prepare and 2 per use.

22
00:01:34.133 --> 00:01:40.767
Equality occurs when 8 + q = 2 + 2q, so q = 6.

23
00:01:40.767 --> 00:01:46.000
At 7 uses, M costs 15 and C 16.

24
00:01:46.000 --> 00:01:53.200
If M's per-use cost were 3 instead, its larger
setup and steeper slope would never recover.

25
00:01:53.200 --> 00:01:58.833
The retained EPFL calculation includes cases that never break even.

26
00:01:58.833 --> 00:02:05.733
A modeled crossing is conditional on stable
costs, valid reuse and the charged lifecycle.

27
00:02:05.733 --> 00:02:11.933
Reload, invalidation, retained memory and extraction can move or remove it.

28
00:02:11.933 --> 00:02:16.654
Do not use a resident model for independently cold requests.

29
00:02:16.654 --> 00:02:16.667
Do not use a resident model for independently cold requests.

30
00:02:16.933 --> 00:02:22.833
If both methods cost three per use and the candidate begins six units behind, no number of

31
00:02:22.833 --> 00:02:26.167
uses closes that gap under this model.

32
00:02:26.167 --> 00:02:33.733
Reuse alone does not guarantee a crossover; the recurring saving must actually exist.

33
00:02:33.733 --> 00:02:39.767
Return to the first model, but suppose each object becomes invalid after two uses.

34
00:02:39.767 --> 00:02:43.400
Each lifetime ends before the fourth-use win.

35
00:02:43.400 --> 00:02:47.167
The next object starts with a new preparation cost.

36
00:02:47.167 --> 00:02:53.933
Adding uses across incompatible versions would invent reuse that never happened.

37
00:02:53.933 --> 00:02:59.200
A break-even estimate is conditional on its costs and lifetime assumptions.

38
00:02:59.200 --> 00:03:03.867
Real measurements can vary, and output delivery can change the model.

39
00:03:03.867 --> 00:03:08.733
Use the algebra to understand what must be measured, then check that the workload actually

40
00:03:08.733 --> 00:03:12.627
provides the required reuse.

41
00:03:12.627 --> 00:03:12.633
provides the required reuse.

42
00:03:12.633 --> 00:03:16.833
State how long the artifact remains valid and resident.

43
00:03:16.833 --> 00:03:25.233
Include reload, invalidation, extraction, and other
required costs before interpreting a modeled crossing.