Authors: Saeed Fathollahzadeh, Kia Teymourian, Chris Jermaine

Paper Name: How Object Model and Its Implementation Impacts Performance of Big Data Systems: A Controlled Experimental Study

Paper Links:

• Link
• PDF

Source Code Info:

• Repository: serialization (https://github.com/fathollahzadeh/serialization)
• Programming Language: C++, Java, Rust
• Packages/Libraries Needed: OpenJDK, Clang++, Rust

Dataset Used:

• Twiter: https://www.cs.utexas.edu/~kiat/datasets/tweets_1M_rows.txt.bz2

Hardware and Software Info: The experiments use both a scale-up, one-socket AMD server, and a scale-out cluster of 7 single-socket Intel servers:

• Scale-up node: AMD EPYC 7302 CPU @ 3.0-3.3 GHz (16 physical/32 virtual cores) with 512KB, 8MB and 128MB L1, L2 and L3 caches, and 128 GB DDR4 RAM (peak performance is 768 GFLOP/s, 183.2 GB/s).The OpenJDK 11 with 120 GB max and initial JVM heap sizes for Java
• Scale-out cluster: 7 nodes, each a single on Google Cloud Platform using N2 Series which have 4 vCPU cores (Intel Cascade), 32 GB RAM, and 750 GB (2× 375 GB) local SSD (Sustained random IOPS limit: read=340,000 and write=180,000). The OpenJDK 11 with 28 GB max and initial JVM heap sizes for Java based implementations.

Setup and Experiments: The entire experimental evaluation can be run via ./runAll.sh, which deletes the results and plots and performs setup, dataset download and preparation, local and distributed experiments, and plotting. However, for a more controlled evaluation, we recommend running the individual steps separately:

    ./run1SetupDependencies.sh
    ./run2SetupSerialization.sh
    ./run3DownloadData.sh
    ./run4PrepareData.sh
    ./run5LocalExperiments.sh # on scale-up node
    ./run5LocalExperimentsExternalSort.sh # on scale-up GC node
    ./run6NetworkExperiments.sh # on scale-out GC cluster
    ./run7LocalMemoryExperiments.sh # on scale-up node

Object Relationships and Ratio Frequency of Tweet Objects (for one million tweet dataset):

Last Update: Apr 26, 2023 (first draft)