Skip to content

Latest commit

 

History

History
161 lines (85 loc) · 7.81 KB

PATCG_2022.8.11_README.md

File metadata and controls

161 lines (85 loc) · 7.81 KB
Raw

PATCG 8/11 Benchmarks

Here are the run commands for the benchmarks we shared of IPA the the 8/11/22 PATCG.

We ran with the 3 party honest majority malicious protocol sy-rep-ring-party.x with the default statistical security of 40 bits. For semi-honest we ran replicated secret sharing protocol ring.sh.

To enable mod 2^k ring computation with custom k, add MOD = -DRING_SIZE=32 to MP-SPDZ/CONFIG.mine and rebuild with make -B -j8 sy-rep-ring-party.x . All of our results are were for k = 32.

All of the MPC code is in the research_prototype/ipa/ directory, and can be accessed by running python ipa within this directory (research_prototype/.) This provides a CLI for both compiling and generating input.

You'll need to fill in the <p0.hostname> value in the commands.

The particular commit we benchmarked was https://github.com/bmcase/raw-ipa/commit/8432970926cecef90da3e2bf6202577950d887d6 in the file vectorized.mpc. The structure of the code now is quite different, and the commands below reflect the current API.

Scaling # rows

2^12 4

  • /usr/bin/time --verbose python ipa compile --numrows_power=12
  • python ipa generate_input --numrows_power=12
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 ipae2e --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 ipae2e --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 ipae2e --hostname <p0.hostname>

2^13 4

  • /usr/bin/time --verbose python ipa compile --numrows_power=13
  • python ipa generate_input --numrows_power=13
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 ipae2e --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 ipae2e --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 ipae2e --hostname <p0.hostname>

2^14 4

  • /usr/bin/time --verbose python ipa compile --numrows_power=14
  • python ipa generate_input --numrows_power=14
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 ipae2e --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 ipae2e --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 ipae2e --hostname <p0.hostname>

2^15 4

  • /usr/bin/time --verbose python ipa compile --numrows_power=15
  • python ipa generate_input --numrows_power=15
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 ipae2e --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 ipae2e --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 ipae2e --hostname <p0.hostname>

2^16, 4

  • /usr/bin/time --verbose python ipa compile --numrows_power=16
  • python ipa generate_input --numrows_power=16
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 ipae2e --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 ipae2e --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 ipae2e --hostname <p0.hostname>

2^17, 4

  • /usr/bin/time --verbose python ipa compile --numrows_power=17
  • python ipa generate_input --numrows_power=17
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 ipae2e --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 ipae2e --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 ipae2e --hostname <p0.hostname>

2^18, 4

  • /usr/bin/time --verbose python ipa compile --numrows_power=18
  • python ipa generate_input --numrows_power=18
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 ipae2e --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 ipae2e --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 ipae2e --hostname <p0.hostname>

2^19, 4

  • /usr/bin/time --verbose python ipa compile --numrows_power=19
  • python ipa generate_input --numrows_power=19
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 ipae2e --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 ipae2e --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 ipae2e --hostname <p0.hostname>

Measured Components Data, 2^15, 16

Arguments order: numrows, numbreakdowsn, n_bits, sort_type, do_sort, do_attribution, do_capping, do_aggregation

We always used sort_type =3 for everything which is the MP-SPDZ library radix sort.

We always used do_capping =2 which is our parallel version of the capping algorithm.

To get the performance of each individual stage we compiled and ran with and without that individual stage and subtracted the results.

2^15, 16, sort only

  • /usr/bin/time --verbose python ipa compile --numrows_power=15 --breakdown_values=16 --skip_attribution --skip_capping --skip_aggregation
  • python ipa generate_input --numrows_power=15 --breakdown_values=16
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 vectorized-15-16-32-3-1-0-0-0 --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 vectorized-15-16-32-3-1-0-0-0 --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 vectorized-15-16-32-3-1-0-0-0 --hostname <p0.hostname>

2^15, 16, sort+attribution

  • /usr/bin/time --verbose python ipa compile --numrows_power=15 --breakdown_values=16 --skip_capping --skip_aggregation
  • python ipa generate_input --numrows_power=15 --breakdown_values=16
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 vectorized-15-16-32-3-1-1-0-0 --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 vectorized-15-16-32-3-1-1-0-0 --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 vectorized-15-16-32-3-1-1-0-0 --hostname <p0.hostname>

2^15, 16, sort+attribution+capping

    • /usr/bin/time --verbose python ipa compile --numrows_power=15 --breakdown_values=16 --skip_aggregation
    • python ipa generate_input --numrows_power=15 --breakdown_values=16
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 vectorized-15-16-32-3-1-1-2-0 --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 vectorized-15-16-32-3-1-1-2-0 --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 vectorized-15-16-32-3-1-1-2-0 --hostname <p0.hostname>

2^15, 16, sort+attribution+capping+aggregation

  • /usr/bin/time --verbose python ipa compile --numrows_power=15 --breakdown_values=16
  • python ipa generate_input --numrows_power=15 --breakdown_values=16
  • P0: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 0 vectorized-15-16-32-3-1-1-2-1 --hostname <p0.hostname>
  • P1: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 1 vectorized-15-16-32-3-1-1-2-1 --hostname <p0.hostname>
  • P2: /home/ec2-user/workspace/MP-SPDZ/Scripts/../sy-rep-ring-party.x --player 2 vectorized-15-16-32-3-1-1-2-1 --hostname <p0.hostname>