throughput

Throughput

How to build

cmake . -B build
cd build
make -j

Prerequisite before running

• Move *.pem and *.sh to your build directory.
• Follow the instructions in README.md in tools directory regarding how to generate materials of 96 bidders for throughput evaluation. And move the outputs of these materials to the build directory.

How to run

Non-private baseline throughput

• On the test machine, run bash run-plain-server.sh $PARALLEL_NUM $DIR_OF_SERVER_OUTPUTS.
  • PARALLEL_NUM is the number of servers you run in parallel, it's recommended that you set it to the number of cores in your machine.
  • DIR_OF_SERVER_OUTPUTS is the directory where the outputs files of each server will be written.
• On the same machine, run bash run-plain-client.sh $PARALLEL_NUM $LOAD_PER_SEC $DIR_OF_CLIENT_OUTPUTS.
  • DIR_OF_CLIENT_OUTPUTS is the directory where the outputs files of each client will be written.
  • LOAD_PER_SEC is the number of requests that each client program sends to the server in one second.
• Run pkill plain-server every time you finish running.

Addax throughput

• On one test machine, run bash run-addax-server.sh $PARALLEL_NUM $DIR_OF_SERVER_OUTPUTS $DIR_WITH_SHARES $SHARE_FILES_NAMES_DIR $ROUND.
  • PARALLEL_NUM is the number of servers you run in parallel, it's recommended that you set it to the number of cores in your machine.
  • DIR_WITH_SHARES is the directory where the shares are stored.
  • SHARE_FILES_NAMES_DIR is the directory that contains the file names of the shares.
  • DIR_OF_SERVER_OUTPUTS is the directory where the outputs files of each server will be written.
  • If you want to run first-price baseline add -q as the last parameter of the command.
• On the same machine, run bash run-addax-client.sh $PARALLEL_NUM $LOAD_PER_SEC $DIR_OF_CLIENT_OUTPUTS $ROUND
  • DIR_OF_CLIENT_OUTPUTS is the directory where the outputs files of each client will be written.
  • LOAD_PER_SEC is the number of requests that each client program sends to the server in one second.
• Run pkill addax-server every time you finish running.

Analyze throughput from the outputs

• To compute the median and 99% latency of all the requests from the client, run python cal-latency.py DIR_OF_CLIENT_OUTPUTS.
• To compute the throughput (i.e., how many tasks are finished within the given time frame) of the server, run python cal-throughput.py DIR_OF_SERVER_OUTPUTS LOAD_PER_SEC.