Journey

This is the repo for our CS736 project, including microbenchmarks, results, figures, and paper.

To build all benchmarks, use make (I recommend also using -j).

Here is a quick list of the contents:

• Microbenchmarks: A bunch of C++ files. The naming convention is bench_, followed by what metric is being measured (cpu, time, mem, ops_per_sec), followed by what operation/subsystem is being benchmarked (kswapd, mmap), followed by what workload is being run (the easiest thing is to read the code to see what pattern is used).

  • bench_cpu_kswapd_anon.cc
  • bench_cpu_kswapd_anon_rand.cc
  • bench_measure_cpu_util_cpu.cc
  • bench_mem_control.cc
  • bench_mem_mmap_16k_cont.cc
  • bench_mem_mmap_16k_cont_rwx_prot_4k_rdonly.cc
  • bench_mem_mmap_16k_cont_rwx_prot_4k_rdonly_touch.cc
  • bench_mem_mmap_16k_cont_touch.cc
  • bench_mem_mmap_4k_rand.cc
  • bench_mem_mmap_anon_rss.cc
  • bench_mem_mmap_anon_rss_rand.cc
  • bench_mem_mmap_frag.cc
  • bench_mem_mmap_stride.cc
  • bench_ops_per_sec_mmap_16k_cont_touch.cc
  • bench_ops_per_sec_mmap_4k_cont_touch.cc
  • bench_time_mmap_anon_rss.cc
  • bench_time_mmap_anon_rss_rand.cc
  • bench_time_per_op_mmap_16k_cont.cc
  • bench_time_per_op_mmap_16k_stride.cc
  • bench_time_per_op_mmap_4k_rand.cc
  • bench_time_per_op_mmap_frag.cc
  • bench_time_per_op_mmap_small_remap_large.cc

• Supporting files

  • pattern.cc, pattern.h - Helpers for producing patterns of memory accesses
  • pinning.h - Helpers for pinning workloads to processor cores
  • rdmemusg.cc, rdmemusg.h - Helpers for reading/computing total system memory usage
  • rdtsc.cc, rdtsc.h - Helpers for measuring time elapsed
  • makefile - Builds all benchmarks
  • offsets.txt - A file containing a bunch of non-repeating random integers

• Scripts

  • do_latency_measurements.py - Runs a bunch of latency-related microbenchmarks

  • run_tests.py - Runs benchmarks specified in a global variable. Be careful when using this, as the python interpretter could affect results for benchmarks that measure global system memory usage.

  • scaling_set_freq.sh - Set the frequency of the processor and scaling governors

  • Scripts for analyzing results and generating graphs. Most of these use matplotlib or scipy. Read the script to see if it suits your purposes.

    • gen_bar.py
    • line_graph.py
    • line_graph_with_control.py
    • time_cdf.py
    • medians.py

• Reports

  • paper.tex - Our final report
  • proposal.tex - Our original idea proposal
  • related_work.tex - Our original related work section
  • references.bib - References for all reports
  • results/ - Final results (raw numbers)
  • figures/ - Images used in paper.tex