Automated Reliability Classification of Queueing Models for - - PowerPoint PPT Presentation

automated reliability classification of queueing models
SMART_READER_LITE
LIVE PREVIEW

Automated Reliability Classification of Queueing Models for - - PowerPoint PPT Presentation

Automated Reliability Classification of Queueing Models for Streaming Computation Jonathan Beard, Cooper Epstein, and Roger Chamberlain SBS Stream Based Supercomputing Lab http://sbs.wustl.edu Work also supported by: 1 Stream Processing


slide-1
SLIDE 1

Automated Reliability Classification of Queueing Models for Streaming Computation

Jonathan Beard, Cooper Epstein, and Roger Chamberlain

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Work also supported by:

1

slide-2
SLIDE 2

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Stream Processing

2

for i←0 through N do a[i] ←(b[i] + c[i]) i++ end do

i++ a,b,c,i i <=N

a[i] ←(b[i] + c[i])

exit

Read b,c

  • ut <- b + c

Write a Traditional Control Flow Streaming

slide-3
SLIDE 3

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Stream Processing

3

Kernel Assigned To Any Compute Resource

slide-4
SLIDE 4

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

RaftLib

C++ Streaming Template Library Auto-parallelizes code Manages resources, buffers, TCP links Automatically Optimized Online

4

software download: http://raftlib.io

slide-5
SLIDE 5

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Modeling Streams as Queues

B C Q1 Q2

A

A B C “Stream” is modeled as a Queue

5

slide-6
SLIDE 6

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Buffer Sizing

6

slide-7
SLIDE 7

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Traditional Service Rate

Counter - In Counter - Out Isolated Compute Kernel

7

V1 V2 s t

slide-8
SLIDE 8

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Example Streaming App

8

Read Matrix Dot Product Dot Product Reduce Dot Product

1 i n

Matrix Multiply

slide-9
SLIDE 9

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Example Streaming App

9

Rabin-Karp String Search

Rolling Hash Read File, Distribute Rolling Hash Rolling Hash Reduce Verify Match Verify Match

1 j i n 1

slide-10
SLIDE 10

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

More Complex Example

10

ID: 0 Name: AppVertex ID: 1 Name: AppVertex ID: 2 Name: AppVertex ID: 10 Name: AppVertex ID: 11 Name: AppVertex ID: 3 Name: AppVertex ID: 4 Name: AppVertex ID: 5 Name: AppVertex ID: 6 Name: AppVertex ID: 7 Name: AppVertex ID: 9 Name: AppVertex ID: 8 Name: AppVertex ID: 16 Name: AppVertex ID: 17 Name: AppVertex ID: 18 Name: AppVertex ID: 12 Name: AppVertex ID: 13 Name: AppVertex ID: 14 Name: AppVertex ID: 15 Name: AppVertex

Too Many Buffers to Manually Model

slide-11
SLIDE 11

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

“Knowledge” Transfer

11

Machine Learning

“Knowledge”

slide-12
SLIDE 12

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Support Vector Machine

12

Can we use a Support Vector Machine (SVM) to classify a model as use or don’t use?

B Q1

A

M/D/1 M/M/1

None

?

slide-13
SLIDE 13

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Features

13

Kernel

slide-14
SLIDE 14

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Monitor Arrangement

14

processor core processor core Kernel Thread Monitor Thread Kernel Thread processor core OS Scheduler Kernel A Kernel B Stream

slide-15
SLIDE 15

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Online Service Rate

15

Rate Shift Here

slide-16
SLIDE 16

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Kernel

16

Kernel

SVM

slide-17
SLIDE 17

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Applying the SVM

17

SVM

B Q1

A

M/D/1 M/M/1 None

slide-18
SLIDE 18

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Training & Evaluation Methodology

> 45k micro-benchmark executions > 1,000 full application executions (2 apps) Multiple hardware types (ARM, x86, PowerPC) Multiple operating systems (Linux, OS X, BSD) Training / Testing

  • 20% of both micro-benchmark and

application data used for training

  • 80% used for testing

18

slide-19
SLIDE 19

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Test App 1

19

Read Matrix Dot Product Dot Product Reduce Dot Product

1 i n

Matrix Multiply

slide-20
SLIDE 20

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Test App 2

20

Rabin-Karp String Search

Rolling Hash Read File, Distribute Rolling Hash Rolling Hash Reduce Verify Match Verify Match

1 j i n 1

slide-21
SLIDE 21

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Correct Classification by M/M/1

21

slide-22
SLIDE 22

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Correct Classification by M/D/1

22

slide-23
SLIDE 23

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Test Classification Rates M/M/1

23

Server Utilization

slide-24
SLIDE 24

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

Test Classification Rates M/D/1

24

Server Utilization

slide-25
SLIDE 25

SBS

Stream Based Supercomputing Lab

http://sbs.wustl.edu

25

RaftLib: http://raftlib.io My Page: http://cse.wustl.edu/~beardj Lab Page: http://sbs.wustl.edu