(ATS6-GS04) Performance Analysis of Accelrys Enterprise Platform 9.0 on IBM’s Scalable IT solution
0 likes588 views
The document presents a performance analysis of the Accelrys Enterprise Platform 9.0 on IBM's scalable IT solutions, focusing on various mapping modules such as Bowtie, Bowtie2, and BWA. It includes test case descriptions, benchmarking results, and scalability analyses showing the elapsed time and performance improvements across different configurations and file systems. Additionally, best practices and architectural insights for maximizing performance in genomic data processing are outlined.
(ATS6-GS04) Performance Analysis of Accelrys Enterprise Platform 9.0 on IBM’s Scalable IT solution
- 1. Performance Analysis of AccelrysPerformance Analysis of Accelrys Enterprise Platform 9.0 on IBMEnterprise Platform 9.0 on IBM’’ss Scalable IT solutionScalable IT solution Kathy Tzeng, PhD Senior Technical Staff Member IBM tzy@us.ibm.com
- 2. © 2013 IBM Corporation Content • Performance of AEP 9.0 – NGS collection: Bowtie, Bowtie2, BWA • Effect of File System on Performance – NGS Collection: Bowtie2 – Chemistry: Energy Minimization • Best Practice – Reference Architecture
- 3. © 2013 IBM Corporation Performance of AEP 9.0
- 4. © 2013 IBM Corporation System Configuration
- 5. © 2013 IBM Corporation Test Case Description • Mapping Module: – Bowtie, Bowtie2 and BWA Mapper in the NGS Collection • Experiment: – SRX000600, Illumina sequencing of Human HapMap individual NA18507 genomic paired-end library – 213 runs, 1.6 billion spots, 117.4 G bases – Input: ~158GB compressed (~600GB uncompressed) FASTQ files – Output: 180 GB BAM files (~600 GB tmp files) • Reference: – Full Human Genome (3GB) build 37.3
- 6. © 2013 IBM Corporation Bowtie Mapping Protocol
- 7. © 2013 IBM Corporation Performance of Bowtie Protocol 303 160 92 65 213 135 125 187 97 53 36 0 50 100 150 200 250 300 350 8 16 32 64 Number of Nodes ElapsedTime(minutes) Bowtie: 1 process/node, 32 threads/process Bowtie: 4 process/node, 8 threads/process Bowtie: 4 process/node, 8 threads/process w/ Splitter on 2.6 GHz iDataPlex with GPFS
- 8. © 2013 IBM Corporation Scalability Analysis of Bowtie Protocol Bowtie benchmark comparative performance 187 97 53 36 1.00 1.93 3.53 5.19 0 20 40 60 80 100 120 140 160 180 200 8.0 16.0 32.0 64.0 Number of Nodes ElapsedTime(seconds) 1 10 RelativeSpeedup Bow tie Runtime Bow tie Speedup Linear Scale on 2.6 GHz iDataPlex with GPFS
- 9. © 2013 IBM Corporation Bowtie 2 Mapping Protocol
- 10. © 2013 IBM Corporation Performance of Bowtie2 protocol 385 189 119 82 301 196 189 254 129 69 48 0 50 100 150 200 250 300 350 400 450 8 16 32 64 Number of Nodes ElapsedTime(minutes) Bowtie2: 1 process/node, 32 threads/process Bowtie2: 4 process/node, 8 threads/process Bowtie2: 4 process/node, 8 threads/process w/ Splitter on 2.6 GHz iDataPlex with GPFS
- 11. © 2013 IBM Corporation Bowtie2 benchmark comparative performance 254 129 69 48 1.00 1.97 3.68 5.29 0 50 100 150 200 250 300 8.0 16.0 32.0 64.0 Number of Nodes ElapsedTime(seconds) 1 10 RelativeSpeedup Bow tie2 Runtime Bow tie2 Speedup Linear Scale Scalability Analysis of Bowtie2 Protocol on 2.6 GHz iDataPlex with GPFS
- 12. © 2013 IBM Corporation BWA Mapping Protocol
- 13. © 2013 IBM Corporation Performance of BWA protocol 305 162 93 74 179 100 75 75 0 50 100 150 200 250 300 350 8 16 32 64 Number of Nodes ElapsedTime(minutes) BWA: 4 process/node, 8 threads/process w/ Splitter BWA: 8 process/node, e threads/process w/ Splitter4 on 2.6 GHz iDataPlex with GPFS
- 14. © 2013 IBM Corporation Scalability Analysis of BWA Protocol BWA benchmark comparative performance 179 100 75 75 1.00 1.79 2.39 2.39 0 20 40 60 80 100 120 140 160 180 200 8.0 16.0 32.0 64.0 Number of Nodes ElapsedTime(seconds) 1 10 RelativeSpeedup BWA Runtime BWA Speedup Linear Scale on 2.6 GHz iDataPlex with GPFS
- 15. © 2013 IBM Corporation Summary of performance of NGS collection Elapsed time for mapping 30x coverage of human genome on 2.6 GHz iDataPlex with GPFS
- 16. © 2013 IBM Corporation Effect of File System on Performance
- 17. © 2013 IBM Corporation File System Description 4 x DCS3700 (240 x 2TB, SAS, 7.2K RPM)/vol/xcae1 DDN SFA12k (280 x 3TB, SAS, 7.2K RPM)/gpfs Storage SystemFile System
- 18. © 2013 IBM Corporation Energy Minimization Test Cases Energy Minimization Protocol with SD Files Energy Minimization Protocol without SD Files
- 19. © 2013 IBM Corporation Energy Minimization Benchmarks 228227No SD file 629356Write SD file /vol/xcae1/gpfsTest Case * Benchmark was run by using 32 processes on single node on 2.6 GHz iDataPlex with GPFS and NFS Elapsed Time in seconds, lower is better
- 20. © 2013 IBM Corporation Bowtie2 Benchmarks 437119Bowtie2 /vol/xcae1/gpfsTest Case on 2.6 GHz iDataPlex with GPFS and NFS Elapsed Time in Minutes, lower is better * Benchmark was run by using 32 nodes
- 21. © 2013 IBM Corporation File Cache Library Improves I/O Performance Without cache library With cache libraryWithout cache libraryWithout cache library With cache libraryWithout cache library Elapsed Time = 1730 min ElapsedTime = 107 min
- 22. © 2013 IBM Corporation Best Practice
- 23. © 2013 IBM Corporation IBM Reference Architecture for AEP
- 24. © 2013 IBM Corporation • Refined / matured for past 13 years • Over 100,000 GPFS licenses worldwide • In academic, commercial and government • Proven scalability for thousands of nodes • Proven capability for greater than 10 PB in single file system • File system performance > 1,500 GB/sec • Amount of archival storage > 500 PB • http://www-03.ibm.com/systems/software/gpfs/ IBM General Parallel File System
- 25. © 2013 IBM Corporation Q & A