CADET

A Systematic Method to Debugging Misconfigurations Using Counterfactual Reasoning

Transcript

2. 2 ▷ ▷ ▷ ▷
3. • Misconfigurations can elicit unexpected interactions between software and hardware • These can result in non-functional faults ◦ Affecting non-functional system properties like latency, throughput, energy consumption, etc. 3 !
6. 6 nvpmodel=MAX-N jetson_clock.sh + set(CUDA_STATIC_RUNTIME OFF) ... + -gencode=arch=compute_62,code=sm_62 ✖ ✖ ✖ ✖ !
8. 8 !
10. 10 !
11. 11 • •
12. 12 ? ?
13. 13 • •
14. 14 • •
15. 15 GPU Growth Swap Mem. Load Latency c1 0.2 2 Gb 10% 1 sec c2 0.5 1 Gb 20% 2 sec cn 1.0 4 Gb 40% 0.1 sec ⋮ ⋮ ⋮ ⋮ ⋮
17. 17 Yes No
18. 18 Yes No
20. 20 • •
21. • In real world cases, this causal graph can be very complex • It may be intractable to reason over the entire graph directly 21 • • •
23. • Compute the Average Causal Effect (ACE) of each pair of neighbors in a path 23 𝐴𝐶𝐸 , = 1 𝑁 ෍ 𝑎,𝑏∈𝑍 𝔼 𝑑𝑜 = 𝑏 − 𝔼 𝑑𝑜 = 𝑎
24. • Sum the ACE of all pairs of adjacent nodes in the path • Rank paths from highest path ACE (PACE) score to the lowest • Use the top K paths for subsequent analysis 24 𝑃𝐴𝐶𝐸 𝑋,𝑍 = 1 𝑁 ෍ 𝑍,𝑋∈𝑃𝑎𝑡ℎ 𝐴𝐶𝐸(𝑍, 𝑋)
25. 25 • •
26. • Counterfactual inference asks “what if” questions about changes to the misconfigurations 26 • • • • •
28. 28 𝑃𝑜𝑡𝑒𝑛𝑡𝑖𝑎𝑙 = 𝑃 ෣ 𝐿𝑎𝑡𝑒𝑛𝑐𝑦 = 𝑙𝑜𝑤. . ෣ 𝑆𝑤𝑎𝑝 = 4 𝐺𝑏, . 𝑆𝑤𝑎𝑝 = 2 𝐺𝑏,. 𝐿𝑎𝑡𝑒𝑛𝑐𝑦 𝑠𝑤𝑎𝑝=2𝐺𝑏 = ℎ𝑖𝑔ℎ,. 𝑈 )
29. 29 Potential = 𝑃 ෟ 𝑜𝑢𝑡𝑐𝑜𝑚𝑒 = 𝑔𝑜𝑜𝑑~ ~𝑐ℎ𝑎𝑛𝑔𝑒,~ 𝑜𝑢𝑡𝑐𝑜𝑚𝑒¬𝑐ℎ𝑎𝑛𝑔𝑒 = 𝑏𝑎𝑑,~¬𝑐ℎ𝑎𝑛𝑔𝑒, ~𝑈 ) Individual Treatment Effect = Potential − Outcome Control = 𝑃 ෟ 𝑜𝑢𝑡𝑐𝑜𝑚𝑒 = 𝑏𝑎𝑑~ ~¬𝑐ℎ𝑎𝑛𝑔𝑒, ~𝑈 )
30. 30 ⋮ 𝐼𝑇𝐸(𝑐ℎ𝑎𝑛𝑔𝑒)
31. 31 • • •
32. 32 • •
34. 3 Embedded real-time stereo estimation Source code ×
35. 3 ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
38. • Finding fixes for performance faults when deployment environment changes 38 • • ?
39. • Finding fixes for performance faults when deployment environment changes 39 • • ?
40. • •