README.md

# CES Adapted slurm_simulator_tools

This repo provides a tweaked and adapted version of BSC-RM's [slurm_simulator_tools](https://github.com/BSC-RM/slurm_simulator)
with user support as seen at [UBCCR's simulator](https://github.com/ubccr-slurm-simulator/slurm_simulator). 

An associated [Docker Image](https://earth.bsc.es/gitlab/mgimenez/docker-ubuntu-ces-slurm-sim) was developed, which is meant to work out of the box. 
Regardless, if you prefer to installit manually, we provide the dependencies.

BSC-CNS - Earth Sciences - Computational Earth Sciences group

Manuel Giménez de Castro Marciani ([manuel.gimenez@bsc.es](mailto:manuel.gimenez@bsc.es))

## Dependencies

This is a comprehensive list of dependencies. Assuming you are installing to ubuntu (debian):

 - make (4.2.1-1.2)
 - automake (1:1.16.1-4ubuntu6)
 - autoconf (2.69-11.1)
 - gcc (4:9.3.0-1ubuntu2)
 - g++ (4:9.3.0-1ubuntu2)
 - libcurl4 (7.68.0-1ubuntu2.18)
 - debianutils (4.9.1)
 - libglib2.0-dev (2.64.6-1~ubuntu20.04.4)
 - libgtk2.0-dev (2.24.32-4ubuntu4)
 - mariadb-server (1:10.3.38-0ubuntu0.20.04.1)
 - mariadb-client (10.3.38-0ubuntu0.20.04.1) 
 - libmariadbd-dev (1:10.3.38-0ubuntu0.20.04.1) 
 - libpam0g-dev (1.3.1-5ubuntu4.6)
 - bzip2 (1.0.8-2)
 - libtool (2.4.6-14)
 - libncurses-dev (6.2-0ubuntu2)
 - libgdm1 (3.36.3-0ubuntu0.20.04.4)
 - libffi7 (3.3-4)
 - zlib1g-dev (1:1.2.11.dfsg-2ubuntu1.5)
 - zlib1g (1:1.2.11.dfsg-2ubuntu1.5)
 - locales (2.31-0ubuntu9.9)
 - python (3.8.2-0ubuntu2)

Moreover, there is one python dependency
 
 - pandas (2.0.1)

## Installation

To install the Simulator:
- Install dependencies
- Run: ./install_slurm_sim.sh
- Restart the machine

## Configuration

This simulator is adapted from [the original](https://github.com/BSC-RM/slurm_simulator) because we add user accounting. 
In order to do so, we need to create and configure the database. Launch your sql server (either mysql or mariadb) and 

```sql
CREATE DATABASE slurm_acct_db;
GRANT USAGE ON *.* to 'root'@'localhost';
GRANT ALL PRIVILEGES on slurm_acct_db.* to 'root'@'localhost';
FLUSH PRIVILEGES; 
```

WARNING: Since this repo is prepared to be ran in a docker image, it is ran by `root`.

## Input

This simulator uses SWF input, which is a [workload format](https://www.cs.huji.ac.il/labs/parallel/workload/swf.html) proposed 
by prof. Dror Feitelson. 

## Output

Script `install/slurm_varios/trace.sh.template` is launched everytime a job has ended in slurm, and it handles the output format.
By the original developers, the output format is:

```
JobId=$JOBID UserId=$UID GroupId=$GID Name=$JOBNAME JobState=$JOBSTATE Partition=$PARTITION TimeLimit=$LIMIT SubmitTime=$SUBMIT StartTime=$START EndTime=$END NodeList=$NODES NodeCnt=$NODECNT ProcCnt=$PROCS WorkDir=$WORK_DIR Backfilled=$BACKFILLED"
```

Feel free to change it however you prefer.

## Running the simulation

Once everything is configured, the whole simulation is handled by the script `run_simulation.py`, which is comprised of

- starting `slurmdbd`
- configuring simulated users
- adapting swf format to binary to feed `sim_mgr` process
- configuring each folder where the simulation will be held

In order to run a simulation run

```bash 
python3 run_simulation.py <path_to_swf_file>
```

## Acknowledgement

If you use this software on a paper, cite the [original authors](https://github.com/BSC-RM/slurm_simulator) of it:

"Evaluating SLURM Simulator with Real-Machine SLURM and Vice Versa", Ana Jokanovic, Marco D'Amico, Julita Corbalan, Performance Modeling, Benchmarking and Simulation of High Performance Computer Systems (PMBS18)At: ACM/IEEE Supercomputing 2018 (SC18), Dallas, TX, USA