TTC 2018 CASE PRESENTATION Quality-based Software-Selection and - - PowerPoint PPT Presentation

▶
ttc 2018 case presentation
SMART_READER_LITE
LIVE PREVIEW

TTC 2018 CASE PRESENTATION Quality-based Software-Selection and - - PowerPoint PPT Presentation

Oct 04, 2023 290 likes •635 views

, TTC 2018 CASE PRESENTATION Quality-based Software-Selection and Hardware-Mapping as a Model T ransformation Problem Sebastian Gtz, Johannes Mey, Rene Schne and Uwe Amann TTC 2018 Case Presentation Slide 1 of 30 , The TTC Case

slide-1
SLIDE 1

,

TTC 2018 CASE PRESENTATION

Quality-based Software-Selection and Hardware-Mapping as a Model T ransformation Problem Sebastian Götz, Johannes Mey, Rene Schöne and Uwe Aßmann

TTC 2018 Case Presentation Slide 1 of 30

slide-2
SLIDE 2

,

The TTC Case

Optimally combine heterogeneous hardware and adaptive software by deriving a solution model from a problem model.

TTC 2018 Case Presentation Slide 2 of 30

slide-3
SLIDE 3

,

Our History of the Case

In the beginning, there was a PhD in 2013:

  • [Götz 2013] Multi-Quality Auto-Tuning by Contract Negotiation

TTC 2018 Case Presentation Slide 3 of 30

slide-4
SLIDE 4

,

Our History of the Case

In the beginning, there was a PhD in 2013:

  • [Götz 2013] Multi-Quality Auto-Tuning by Contract Negotiation

which was improved by faster intermediate model generation in 2016:

  • [Schöne et al. 2016] Incremental Runtime-Generation of Optimisation

Problems Using RAG-Controlled Rewriting

TTC 2018 Case Presentation Slide 3 of 30

slide-5
SLIDE 5

,

Our History of the Case

In the beginning, there was a PhD in 2013:

  • [Götz 2013] Multi-Quality Auto-Tuning by Contract Negotiation

which was improved by faster intermediate model generation in 2016:

  • [Schöne et al. 2016] Incremental Runtime-Generation of Optimisation

Problems Using RAG-Controlled Rewriting which was still a bit slow, so now there is

  • TTC 2018

TTC 2018 Case Presentation Slide 3 of 30

slide-6
SLIDE 6

,

The Problem

TTC 2018 Case Presentation Slide 4 of 30

slide-7
SLIDE 7

,

Problem 1: “Software Selection”

  • Software model:

– Software component specifications:

  • functionality

– Implementations of component specs:

  • provide non-functional properties
  • require other components

Selection T ask

  • Fulfill requests

– chose implementations – ensure non-functional requirements

  • Solution Part 1: Trees of assignments

TTC 2018 Case Presentation Slide 5 of 30

slide-8
SLIDE 8

,

Problem 2: “Hardware Mapping”

  • Hardware model

– Resources with sub-resources and properties

  • Contracts

– Implementations specify resource requirements

Resource Allocation T ask

  • Map assignments to hardware

– ensure resource requirements

  • Solution Part 2: Resource mapping

TTC 2018 Case Presentation Slide 6 of 30

slide-9
SLIDE 9

,

Problem 3: “Quality-Based”

  • Contracts

– Implementations provide non-functional properties depending on hardware

Optimization task

  • Optimize aggregated non-functional property of system

– Here: minimize energy

  • Solution Part 3: Assignments + mapping with minimal energy

TTC 2018 Case Presentation Slide 7 of 30

slide-10
SLIDE 10

,

The Models in Detail

  • Model: two grammars with overlay edges and connecting references

– Problem model:

  • software and hardware part

– Solution model:

  • tree of dependent assignments
  • Grammar?

– Reference Attribute Grammar: efficient analysis – Parser available – Simple solution within model

TTC 2018 Case Presentation Slide 8 of 30

slide-11
SLIDE 11

,

The Models in Detail

Components

TTC 2018 Case Presentation Slide 9 of 30

slide-12
SLIDE 12

,

The Models in Detail

Solution Part 1: Implementation Selection

TTC 2018 Case Presentation Slide 10 of 30

slide-13
SLIDE 13

,

The Models in Detail

Solution Part 2: Hardware Mapping

TTC 2018 Case Presentation Slide 11 of 30

slide-14
SLIDE 14

,

The Models in Detail

Solution Part 3: Optimization

TTC 2018 Case Presentation Slide 12 of 30

Valid: Optimal:

slide-15
SLIDE 15

,

The Models in Detail

Solution Part 3: Optimization

TTC 2018 Case Presentation Slide 13 of 30

Valid: Optimal:

slide-16
SLIDE 16

,

Task and Solutions

TTC 2018 Case Presentation Slide 14 of 30

slide-17
SLIDE 17

,

Case Scenarios

  • Five sizes:

– minimal, small, medium, large, huge

  • Three types:

– standard – more hardware components – more (complex) software components

  • Flexible scenario generator:

– 10 parameters for software/hardware config – Fixed hardware types, and software properties – Flexible shape of sotware model and solution tree

TTC 2018 Case Presentation Slide 15 of 30

slide-18
SLIDE 18

,

Case Scenarios

ID Requests Impl’s Resources Scenario 1 1 1 minimal 1 1 6 5 small 2 1 6 15 small-hw 3 1 62 47 small-sw 4 15 30 68 medium 5 15 30 225 medium-hw 6 10 155 465 medium-sw 7 20 60 90 large 8 20 60 300 large-hw 9 20 310 930 large-sw 10 50 150 225 huge 11 50 150 750 huge-hw 12 50 620 2325 huge-sw

TTC 2018 Case Presentation Slide 16 of 30

slide-19
SLIDE 19

,

A Simple Attribute Grammar Reference Solution

  • Simple reference implementation

– Based on reference attribute grammar – Iterator over model – Some pruning

  • Performance:

– Almost full state space exploration – Encouraging for TTC partitipants – Always finds optimal solution ...eventually

TTC 2018 Case Presentation Slide 17 of 30

slide-20
SLIDE 20

,

Evaluation criteria

Solution time Time to compute a valid solution Solution quality: Validity of solution + Quality of found objective value Scalability: Largest scenario for which a valid solution can be found

TTC 2018 Case Presentation Slide 18 of 30

slide-21
SLIDE 21

,

Measurement results

= valid and in time = valid, but timeout = invalid = optimal (if known from ILP solver) Scenario ACO EMFeR ILP (direct/ext) Simple 0 trivial 6 194 24 / 21 1 1 small 8 / 212 37 / 40 6 2 small-hw 11 240 44 / 61 8 3 small-sw 451 7min52s 377 / 572 15min 4 medium 1min33 / 8min22s 8min28s / 15min 5 medium-hw 4min48s 11min15s 15min / 15min 6 medium-sw 15min 11min15s 15min 15min

TTC 2018 Case Presentation Slide 19 of 30

slide-22
SLIDE 22

,

Some Observations

  • ACO sometimes returns invalid solutions
  • ILP direct much better than ILP external
  • EMFeR for scenarios 3-6 aborts search before timeout
  • Simple either is fastest and optimal, or runs into timeout

TTC 2018 Case Presentation Slide 20 of 30

slide-23
SLIDE 23

,

References

[Götz 2013] Götz, Sebastian. “Multi-Quality Auto-Tuning by Contract Negotiation. ” PhD Thesis, Technische Universität Dresden, 2013. http://nbn-resolving.de/urn:nbn:de:bsz:14-qucosa-119938. [Schöne et al. 2016] Schöne, René, Sebastian Götz, Uwe Aßmann, and Christoff Bürger. “Incremental Runtime-Generation of Optimisation Problems Using RAG-Controlled Rewriting. ” In Proceedings of the 11th International Workshop on Models@run.Time. Saint-Malo: ceur, 2016. http://ceur-ws.org/Vol-1742/.

TTC 2018 Case Presentation Slide 21 of 30

slide-24
SLIDE 24

, TTC 2018 Case Presentation Slide 22 of 30

slide-25
SLIDE 25

,

Backup

TTC 2018 Case Presentation Slide 23 of 30

slide-26
SLIDE 26

,

Questions to the Audience

  • Accessibility of the benchmark?
  • Explanation of the case clear enough?
  • How complex was the problem (compared to previous years)?
  • Anything missing or improvable in the benchmark framework?

TTC 2018 Case Presentation Slide 24 of 30

slide-27
SLIDE 27

,

Grammar Hardware

TTC 2018 Case Presentation Slide 25 of 30

slide-28
SLIDE 28

,

Grammar Expression

TTC 2018 Case Presentation Slide 26 of 30

slide-29
SLIDE 29

,

Grammar Software

TTC 2018 Case Presentation Slide 27 of 30

slide-30
SLIDE 30

,

Grammar General

TTC 2018 Case Presentation Slide 28 of 30

slide-31
SLIDE 31

,

Grammar Solution

TTC 2018 Case Presentation Slide 29 of 30

slide-32
SLIDE 32

, TTC 2018 Case Presentation Slide 30 of 30