TCSA: Efficient Localization of Busy-Wait Synchronization Bugs for Latency-Critical Applications

This repository is the implementation of TCSA.

Usage

Performance Data Collection

We have implemented a performance data collection script in the collector folder to automate the collection of performance data by setting parameters for sampling. The sampling parameters include: start time, end time, sampling frequency, and sampling period of performance data sampling.

There are two ways to realize automated sampling:

(1) Set the above parameters in recorder.bash to execute automated sampling and immediately execute recorder.bash.

(2) Setting a timer (e.g., cron) to execute the recorder.bash script file at regular intervals.

Let's take cron as an example to illustrate (2).

First edit the cron task list with crontab -e.
Then add the timed tasks you want to execute. Example: 0 3 * * * * /path/recorder.bash means "Execute the recorder.bash script under the /path path at 3:00 a.m. every day".

After recorder.bash is executed, the performance data is stored in the current directory.

The converter.bash script converts all perfX.data into perfX.txt files.

Of course, it is also possible to perform the performance data collection and conversion process manually.

Localization with TCSA

We automate the location of busy-wait synchronization performance bugs by executing TCSA as shown below. We need to specify the path to the performance data to be analyzed and specify the path to where the result files of the automated location are stored.

Use the command as follows:

python main.py <file_path> <output_path> [direction] [degrees_of_freedom] [threshold]

file_path : The path of the collected function call stack data, include the file name.

output_path : The path to store the output results.

direction : The direction of identification of consecutive identical call stacks. 1 means top-down, 0 means bottom-up.

degrees_of_freedom : The depth when judging consecutive identical events for threads.

threshold : Set the threshold value for the number of durations. The default -1 means use the average value.

Requirements

The specific environment requirements for python are in the requirements.txt file.

python3
Linux perf

Example

People

System Optimization Lab, East China Normal University (SOLE)

Contact Information

If you have any questions or suggestions, please contact Ning Li via ningli@stu#DOTecnu#DOTedu.cn.

Repository Special Description

origin: https://jihulab.com/solecnu/tcsa

mirror: https://github.com/MercuryLc/TCSA

solecnugit/TCSA