Tweet
Tweet

Slide 1

Slide 1 text

CAPABILITY-BASED SCHEDULING OF SCIENTIFIC WORKFLOWS IN THE CLOUD MICHEL KRÄMER

Slide 2

Slide 2 text

THAT’S ME

Slide 3

Slide 3 text

Data processing requirements Very large data sets Heterogeneous services Automated data processing

Slide 4

Slide 4 text

Data processing requirements Very large data sets Heterogeneous services Automated data processing

Slide 5

Slide 5 text

Data processing requirements Very large data sets Heterogeneous services Automated data processing

Slide 6

Slide 6 text

Data processing requirements Very large data sets Heterogeneous services Automated data processing

Slide 7

Slide 7 text

A B D E C Cloud-based Scienti ic Work lows Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows A B D E C Scienti ic Work lows Automated data processing Independent services Distributed environments

Slide 8

Slide 8 text

A B D E C Cloud-based Scienti ic Work lows Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows A B D E C Scienti ic Work lows Automated data processing Independent services Distributed environments

Slide 9

Slide 9 text

A B D E C Cloud-based Scienti ic Work lows Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows A B D E C Scienti ic Work lows Automated data processing Independent services Distributed environments

Slide 10

Slide 10 text

A B D E C Cloud-based Scienti ic Work lows Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows A B D E C Scienti ic Work lows Automated data processing Independent services Distributed environments

Slide 11

Slide 11 text

A B D E C Cloud-based Scienti ic Work lows Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows A B D E C Scienti ic Work lows Automated data processing Independent services Distributed environments

Slide 12

Slide 12 text

A B D E C Cloud-based Scienti ic Work lows Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows A B D E C Scienti ic Work lows Automated data processing Independent services Distributed environments

Slide 13

Slide 13 text

A B D E C Cloud-based Scienti ic Work lows Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows A B D E C Scienti ic Work lows Automated data processing Independent services Distributed environments

Slide 14

Slide 14 text

Challenges of cloud-based work low management Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows

Slide 15

Slide 15 text

A B D E C Cloud-based Scienti ic Work lows Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows Challenges of cloud-based work low management Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows

Slide 16

Slide 16 text

A B D E C Cloud-based Scienti ic Work lows Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows Challenges of cloud-based work low management Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows

Slide 17

Slide 17 text

Challenges of cloud-based work low management Dynamic environment Scalability Fault tolerance Deelman et al. (2018). The future of scienti ic work lows

Slide 18

Slide 18 text

How to schedule heterogeneous processing services?

Slide 19

Slide 19 text

Naïve scheduling Service instances Virtual Machines ... ... A VM1 VM2 VM3 VM4 VMn Docker A Docker B GPU C In-memory C In-memory D Docker GPU

Slide 20

Slide 20 text

Naïve scheduling Service instances Virtual Machines ... ... A VM1 VM2 VM3 VM4 VMn Docker A Docker B GPU C In-memory C In-memory D Docker GPU

Slide 21

Slide 21 text

Naïve scheduling Service instances Virtual Machines ... ... A VM1 VM2 VM3 VM4 VMn Docker Docker GPU In-memory ... Docker GPU In-memory ... Docker GPU In-memory ... Docker GPU In-memory ... Docker GPU In-memory ... A Docker B GPU C In-memory C In-memory D Docker GPU

Slide 22

Slide 22 text

Contribution Dynamic environment Scalability Fault tolerance

Slide 23

Slide 23 text

Contribution Dynamic environment Capability-based scheduling Scalability Fault tolerance

Slide 24

Slide 24 text

Contribution Dynamic environment Capability-based scheduling Scalability Fault tolerance Software architecture + Algorithm

Slide 25

Slide 25 text

Software architecture

Slide 26

Slide 26 text

Overview HTTP server Controller Scheduler Instance 1 Instance n Agent ... Database Event bus Cloud manager H C S A M

Slide 27

Slide 27 text

Overview HTTP server Controller Scheduler Instance 1 Instance n Agent ... Database Event bus Cloud manager H C S A M

Slide 28

Slide 28 text

Overview HTTP server Controller Scheduler Instance 1 Instance n Agent ... Database Event bus Cloud manager H C S A M

Slide 29

Slide 29 text

Overview HTTP server Controller Instance 1 Instance n Agent ... Database Event bus Cloud manager H C S A M Scheduler

Slide 30

Slide 30 text

Overview HTTP server Scheduler Instance 1 Instance n Agent ... Database Event bus Cloud manager H C S A M Controller

Slide 31

Slide 31 text

Overview HTTP server Controller Scheduler Instance 1 Instance n Agent ... Database Event bus Cloud manager H C S A M

Slide 32

Slide 32 text

Overview HTTP server Controller Scheduler Instance 1 Instance n Agent ... Database Event bus Cloud manager H C S A M

Slide 33

Slide 33 text

Overview HTTP server Controller Scheduler Instance 1 Instance n Agent ... Database Event bus Cloud manager H C S A M

Slide 34

Slide 34 text

Overview HTTP server Controller Scheduler Instance n Agent ... Database Event bus Cloud manager H C S A M Instance 1

Slide 35

Slide 35 text

A B D E C Controller Read work low from database Generate new process chains Save process chains into database Wait for results

Slide 36

Slide 36 text

A B D E C Controller Read work low from database Generate new process chains Save process chains into database Wait for results

Slide 37

Slide 37 text

A B D E C Controller Read work low from database Generate new process chains Save process chains into database Wait for results

Slide 38

Slide 38 text

A B D E C Controller Read work low from database Generate new process chains Save process chains into database Wait for results

Slide 39

Slide 39 text

A B D E C Controller Read work low from database Generate new process chains Save process chains into database Wait for results

Slide 40

Slide 40 text

A B D E C Controller Read work low from database Generate new process chains Save process chains into database Wait for results

Slide 41

Slide 41 text

A B D E C Controller Read work low from database Generate new process chains Save process chains into database Wait for results

Slide 42

Slide 42 text

A B D E C Controller Read work low from database Generate new process chains Save process chains into database Wait for results

Slide 43

Slide 43 text

A B D E C Controller Read work low from database Generate new process chains Save process chains into database Wait for results

Slide 44

Slide 44 text

Scheduling algorithm

Slide 45

Slide 45 text

500 process chains Scheduler PC1 Docker A1 C++ PC2 Docker PCi Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... A2 (busy) Docker 2 agents

Slide 46

Slide 46 text

500 process chains Scheduler distinct required capability sets PC1 Docker A1 C++ PC2 Docker PCi Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... 2 agents A2 (busy) Docker

Slide 47

Slide 47 text

500 process chains 2 agents Scheduler PC1 Docker A1 C++ A2 (busy) Docker PC2 Docker PCi Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Python TensorFlow GPU

Slide 48

Slide 48 text

500 process chains 2 agents Scheduler PC1 Docker A1 C++ A2 (busy) Docker PC2 Docker PCi Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Python TensorFlow GPU

Slide 49

Slide 49 text

500 process chains 2 agents Scheduler PC1 Docker A1 C++ A2 (busy) Docker PC2 Docker PCi Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... busy ??? Docker Python TensorFlow GPU

Slide 50

Slide 50 text

500 process chains 2 agents Scheduler Cloud Manager PC1 Docker A1 C++ A2 (busy) Docker PC2 Docker PCi Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Python TensorFlow GPU request agents

Slide 51

Slide 51 text

500 process chains 2 agents Scheduler Cloud Manager PC1 Docker A1 C++ A2 (busy) Docker PC2 Docker PCi Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... create Docker Python TensorFlow GPU

Slide 52

Slide 52 text

500 process chains 5 agents Scheduler Cloud Manager PC1 Docker A1 C++ A3 Docker A5 GPU Docker A2 (busy) Docker PC2 Docker PCi Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... create Docker Python TensorFlow GPU

Slide 53

Slide 53 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 Docker A2 (busy) Docker PC2 Docker PCi Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... distinct required capability sets A5 GPU Docker

Slide 54

Slide 54 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 Docker A2 (busy) Docker PC2 Docker PCi Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Python TensorFlow GPU A5 GPU Docker

Slide 55

Slide 55 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 Docker A2 (busy) Docker PC2 Docker PCi Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Python TensorFlow GPU A5 GPU Docker

Slide 56

Slide 56 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 Docker A2 (busy) Docker PC2 Docker PCi Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Python TensorFlow GPU busy OK OK OK ??? A5 GPU Docker

Slide 57

Slide 57 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 Docker A3 Docker A2 (busy) Docker PC2 Docker PCi Python TensorFlow A4 Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Python TensorFlow GPU A5 GPU Docker A5 GPU Docker A5 GPU Docker

Slide 58

Slide 58 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 Docker A3 Docker A2 (busy) Docker PC2 Docker PCi Python TensorFlow A4 Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Python TensorFlow GPU A5 GPU Docker A5 GPU Docker

Slide 59

Slide 59 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 Docker A3 Docker A2 (busy) Docker PC2 Docker PCi Python TensorFlow A4 Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Docker Python TensorFlow GPU A5 GPU Docker A5 GPU Docker fetch process chain

Slide 60

Slide 60 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 Docker A3 Docker A2 (busy) Docker PC2 Docker PCi Python TensorFlow A4 Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Python TensorFlow GPU A5 GPU Docker A5 GPU Docker

Slide 61

Slide 61 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 (busy) Docker A3 Docker A2 (busy) Docker PC2 Docker PCi Python TensorFlow A4 Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Docker Python TensorFlow GPU A5 GPU Docker A5 GPU Docker

Slide 62

Slide 62 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 (busy) Docker A2 (busy) Docker PCi Python TensorFlow A4 (busy) Python TensorFlow A4 Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... Python TensorFlow GPU A5 GPU Docker A5 GPU Docker

Slide 63

Slide 63 text

500 process chains 5 agents Scheduler PC1 Docker A1 C++ A3 (busy) Docker A2 (busy) Docker A4 (busy) Python TensorFlow PCj+1 Docker ... ... PCj GPU PCi+1 GPU ... GPU A5 (busy) GPU Docker A5 GPU Docker

Slide 64

Slide 64 text

500 process chains 5 agents Scheduler Repeat PC1 Docker A1 C++ A3 (busy) Docker A2 (busy) Docker A4 (busy) Python TensorFlow PCj+1 Docker ... ... PCj GPU ... A5 (busy) GPU Docker

Slide 65

Slide 65 text

Evaluation

Slide 66

Slide 66 text

Contribution Dynamic environment Capability-based scheduling Scalability Fault tolerance Software architecture + Algorithm

Slide 67

Slide 67 text

Experiment 1 Capability-based scheduling 100 process chains 4 distinct capability sets Correct allocation R1 R2 R3 R4 R3+R4 Process chain Start End Agent killed Fault

Slide 68

Slide 68 text

Experiment 2 Dynamic environment 1000 process chains 1 agent at the beginning 8 agents at the end R1 R2 R3 R4 R3+R4 Process chain Start End Agent killed Fault

Slide 69

Slide 69 text

Experiment 3 Scalability (process chains) 150.000 process chains up to 8 agents Load managed well R1 R2 R3 R4 R3+R4 Process chain Start End Agent killed Fault

Slide 70

Slide 70 text

Experiment 4 Fault tolerance 1000 process chains Agents randomly killed Successful recovery

Slide 71

Slide 71 text

Implementation of the software architecture and algorithm Open Source https://steep-wms.github.io/ Steep

Slide 72

Slide 72 text

Thanks for listening! MICHEL KRÄMER Fraunhofer IGD, Germany [email protected] github.com/michel-kraemer steep-wms.github.io Icons by Freepik from www. laticon.com