stack-unwind

Stack unwinding Project repo for EECS 582 Fall 2020

Getting Consistent Results

We highly recommend following this blog post to help make your testing results more accurate.

Removing the .eh_frame and .eh_frame_hdr sections from your binary

sudo objcopy --remove-section .eh_frame --remove-section .eh_frame_hdr [binary]

YCSB

For benchmarking with YCSB, please check out ycsb.md.

Specific Program Notes

Redis

• The redis-server binary is the program to run Redis. When measuring executable sizes, this is the binary.
• The redis-cli program can be used as a command-line client.
• Compile flags should be changed in src/Makefile.

Memcached

• Use ./configure to add compile flags
• echo "flush_all" | nc localhost 11211 should work as your 'clear' command in the YCSB script.

MongoDB

• mongo ycsb --eval "db.dropDatabase()" should drop the YCSB database for use in the YCSB script.
• I just used the install-mongod and install-mongo targets to get the server and CLI client, respectively; you shouldn't need to build any other targets.

Clang

• Follow this guide to build clang properly.

GCC

• Refer to Tanvir's "C++ systems and their benchmarks of performance evaluation" on how to build GCC from source.
• Before running ./configure, modify it to add the compile flags you need.

RocksDB

• Follow this guide to compile RocksDB
• Edit the compile flags by editing ./Makefile, and use the PREFIX to mark the installation directory.
• Build only the static library via make static_lib, as other make commands install other, unnecessary tools.
• To better compare stack-unwinding performance impacts on RocksDB against other NoSQL database applications, we recommend using the YCSB benchmark for RocksDB as opposed to its own benchmarking program.

FFmpeg

• Download and unzip vbench.
• In the top directory, run export VBENCH_ROOT=$(pwd)
• Run ./install-reference-ffmpeg.sh for vbench to download all the repositories necessary to build FFmpeg.
• For each new build of FFmpeg, delete or rename the existing ./bin/ffmpeg, modify ./ffmpeg_sources/ffmpeg/configure to add compile flags, and then re-run ./install-reference-ffmpeg.sh
• To create a benchmark, modify the "perform transcoding" section of ./code/reference.py to include a for-loop and a filter for whichever video files (found in ./crf0 and ./crf18) you would like to transcode.
• Then, run the benchmark script with python code/reference.py scenario where scenario can be any of the following: live, upload, platform, popular, or vod