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 for i ← 0 through N do a[i] ← (b[i] + c[i]) i++ end do Traditional Control Flow Streaming a,b,c,i Read b,c exit i <=N out <- b + c a[i] ← (b[i] + c[i]) SBS i++ Write a Stream Based Supercomputing Lab http://sbs.wustl.edu 2
Stream Processing Kernel Assigned To Any Compute Resource SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 3
RaftLib C++ Streaming Template Library Auto-parallelizes code Manages resources, buffers, TCP links Automatically Optimized Online software download: http://raftlib.io SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 4
Modeling Streams as Queues A B C “Stream” is modeled as a Queue A B Q2 C Q1 SBS Stream Based 5 Supercomputing Lab http://sbs.wustl.edu
Buffer Sizing SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 6
Traditional Service Rate V1 V2 s t Isolated Compute Kernel Counter - In Counter - Out SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 7
Example Streaming App Dot Product 1 Read Dot Reduce Matrix Product i Dot Product n Matrix Multiply SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 8
Example Streaming App Rolling Hash 1 Verify Match 1 Read File, Rolling Reduce Distribute Hash i Verify Match j Rolling Hash n Rabin-Karp String Search SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 9
More Complex Example ID: 0 Name: AppVertex ID: 1 Name: AppVertex ID: 2 Name: AppVertex ID: 5 ID: 6 ID: 4 ID: 3 Too Many Buffers to Manually Model Name: AppVertex Name: AppVertex Name: AppVertex Name: AppVertex ID: 8 ID: 7 ID: 10 Name: AppVertex Name: AppVertex Name: AppVertex ID: 16 ID: 9 ID: 11 Name: AppVertex Name: AppVertex Name: AppVertex ID: 14 ID: 13 ID: 15 ID: 12 Name: AppVertex Name: AppVertex Name: AppVertex Name: AppVertex ID: 17 ID: 18 Name: AppVertex Name: AppVertex SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 10
“Knowledge” Transfer Machine “Knowledge” Learning SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 11
Support Vector Machine Can we use a Support Vector Machine (SVM) to classify a model as use or don’t use? None M/D/1 M/M/1 ? A B Q1 SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 12
Features Kernel SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 13
Monitor Arrangement Kernel A Stream Kernel B Kernel Thread Kernel Thread Monitor Thread OS Scheduler processor processor processor SBS core core core Stream Based Supercomputing Lab http://sbs.wustl.edu 14
Online Service Rate Rate Shift Here SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 15
Kernel SVM Kernel SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 16
Applying the SVM M/D/1 M/M/1 None SVM A B Q1 SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 17
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 SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 18
Test App 1 Dot Product 1 Read Dot Reduce Matrix Product i Dot Product n Matrix Multiply SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 19
Test App 2 Rolling Hash 1 Verify Match 1 Read File, Rolling Reduce Distribute Hash i Verify Match j Rolling Hash n Rabin-Karp String Search SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 20
Correct Classification by M/M/1 SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 21
Correct Classification by M/D/1 SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 22
Test Classification Rates M/M/1 Server Utilization SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 23
Test Classification Rates M/D/1 Server Utilization SBS Stream Based Supercomputing Lab http://sbs.wustl.edu 24
RaftLib: http://raftlib.io My Page: http://cse.wustl.edu/~beardj Lab Page: http://sbs.wustl.edu SBS Stream Based 25 Supercomputing Lab http://sbs.wustl.edu
Recommend
More recommend