Data center network communication features for large data applications architecture research
Research Projects:
Hadoop、Spark and Storm big data processing platform benchmark survey.
Using profiling tools and algorithm analysis, extract microBenchmark (ex. sorting, grep, shuffle), obtain microBenchmark call graph.
Foreach microBenchmark, analysis communication features, structure communication model and matching test.
According to the microBenchmark call graph, combine communication models, fitting each benchmark.
Blend Benchmarks, build DataCenter network communication simulator.
Deploy communication simulator on interconnection simulator, evaluate new network topology.
Main Responsibility:
For Hadoop, Spark and Storm. Finish benchmark analysis and profiling.
Extract microBenchmark, using tcpdump measure network flow.
Pitching pile on Hadoop(/Spark/Storm), achieve stage and data size of network production.
Set up Benchmarks’ module.
Run communication simulator on interconnection simulator.
GeoEast system-a typical processing module performance optimization
Research Projects:
PetroChina geological data prestack noise algorithm design.
Multicore optimization on E5-2658A;
Coprocessor optimization on Intel Xeon Phi.
Main Responsibility:
Fortran algorithm realization, c language release;
Carry out 11.3X speed-up on CPU.
Increase 26.5X speed-up on MIC
Deconvolution microscope performance optimization
Research Projects:
The Brain cell image pretreatment and convolution filtering.
Multicore optimization on E5-2658A.
Coprocessor optimization on Intel Xeon Phi.
Main Responsibility:
Matlab algorithm realization, c language release;
Carry out 35.7X speed-up on CPU.
Increase 42.9X speed-up on MIC