This project provides a conduit for routing High-Throughput Computing (HTC) jobs from TACC rsources to a BOINC server, and from there, to the BOINC clients (which run on volunteered hardware resources and VMs in the cloud). The project is funded through NSF Award #1664022.
This repository includes the following:
- Front-end code for the BOINC@TACC website.
- Scripts and API that run on the BOINC server for pulling job requests from TACC resources and pushing the requests to the BOINC clients.
- A back-end script that can be run on TACC resources for determining if a job can be run on BOINC clients, or whether it should be run on TACC resources only.
- A system for generating docker images from source code.
- A mechanism for running BOINC jobs directly inside a Docker container instead of running them in VirtualBox. This is useful for institutional donors who may have volunteered VMs in the cloud.
The project leverages the boinc-server-docker software developed by Marius Millea
Following folders are available in this repository:
- user-interface: Creates a web interface to allow researchers to submit and track jobs.
- api: Designed for large job batches.
- applications: Dockerized applications for running on BOINC clients.
- stampede2-backend: A command-line interface for submitting BOINC jobs from the Stampede2 supercomputer at TACC. This script can be run on any Linux system that has the SLURM scheduler available.
To execute the jobs submitted through the BOINC server, that is, to become a BOINC client, download BOINC and Virtualbox if not installed already. Then, add the project using the following (where SERVER_IP is the IP address of the BOINC server/project website):
http://SERVER_IP/boincserver
The BOINC server will automatically recognize your system as a volunteer and will submit jobs to it when they are available. This project was previously named as TACC-2-BOINC and hence, had a different project URL. Since July 30th 2018, the project is called BOINC@TACC, and the following URL can be used for adding the project to the BOINC client: http://boinc.tacc.utexas.edu
- Install Docker and Docker-compose in their most recent versions
Note: Both docker and docker-compose require sudo access, to avoid this problem:
sudo groupadd docker
sudo gpasswd -a $USER docker
# Log out/in to activate these changes
- Follow the installation instructions provided by boinc-server-docker
- Modify the docker-compose.yml to make it similar to the one provided in this repository. This updated compose file opens the ports necessary for API usage. The only changes needed are in the ports for mysql and apache containers.
- Most likely, this server will be installed on a cloud server. If that is the case, do not compose up directly. It is necessary to specify the IP of the system using the syntax below.
- If not using root, it is possible that an error appears about Docker not being able to connect to the daemon. If this is the case, use sudo to obtain root access, and use sudo for all docker and/or docker-compose commands.
git clone https://github.com/marius311/boinc-server-docker.git
cd boinc-server-docker
docker-compose pull
# Enter the changes on the docker-compose
curl https://raw.githubusercontent.com/ritua2/BOINCatTACC/master/docker-compose.yml > docker-compose.yml
URL_BASE=http://IP_ADDRESS docker-compose up -d
-
Wait a few seconds until the make_project container sets up the BOINC server and its mysql database This process is complete when a webpage is accessible on SERVER_IP/boincserver
-
Set-up basic administrative credentials
- Execute docker ps do obtain all the data about the BOINC containers. If the setup is done correctly, there is an apache container.
- Log into the apache container by using docker exec -it {apache container ID} bash
- Follow the guide in BOINC HTMLOps to set -up an administrative account with .htpasswd
- .htpasswd is located in /home/boincadm/project/html/ops/
- Login into the administrative BOINC page at SERVER_IP/boincserver_ops with the username and password provided above
# Find the docker container
docker ps
docker exec -it boincserverdocker_apache_1 bash
cd /home/boincadm/project/html/ops/
htpasswd -c .htpasswd $NEWUSERNAME
- Clone this repository
- If git is not installed, do it by executing: apt-get update && apt-get install git-core -y
- If not already there, cd /home/boincadm/project
- Clone via: git clone https://github.com/ritua2/BOINCatTACC
- cd BOINCatTACC
cd /home/boincadm/project
git clone https://github.com/ritua2/BOINCatTACC
cd BOINCatTACC
-
Establish user email credentials
- Run the appropriate bash script to set up the email and password
- The basic set-up only allows for gmail, for all others, modify both ./api/preprocessing.py and ./email_assimilator.py
- Be sure to have input the correct name and password, to change them again, change /root/.bashrc
- The same applies to the Docker, MIDAS credentials - MIDAS is a software component for creating docker images from source-code
- Run the external Reef set-up to connect to an external container. The process is entirely automated and designed so that user files are stored in a different system.
Do:
source password_credentials.sh
-
Setup Reef in an exteneral container
- Reef is the storage container where all user files (results or not) are stored
- Create IP following these instructions on a system with the above IP and password provided above
- Note: Reef can be setup on any server as long as its IP is accessible from the main server
-
Run the setup file
- It will install all the necessary packages, python libraries, set-up the internal Redis database, properly locate the files, set-up the APIs, Reef cloud storage, and automatic job processing
- The set-up file will also automatically prompt to enter the credentials for the email. Use caution, since an error would require to manually fix the /root/.bashrc file
cd /home/boincadm/project/BOINCatTACC
bash red_setup.sh
-
Create organization accounts
- A new token must be assigned to each new user (researcher who wants to submit BOINC jobs)
- There are 2 options on assigning tokens:
- Create one automatically through create_token.py (deprecated)
- Create organization accounts
- Organization accounts are preferred since they allow users to sign up in an easier way
- Furthermore, users signing through this way will not have to provide further email authorization
- To create an organization, following the instructions below and enter inputs accordingly
- An automatic Reef account will be created for each user signing up through this method
- Organizations themselves are required to submit jobs using the Stampede2 backend
- Only one organization is needed. However, more can be created in order to separate and differentiate users
- For more instructions on supplying tokens through organizations, check the API documentation
- To do this:
cd /home/boincadm/project/html/user/token_data
python3 create_organization.py
-
Assign VolCon credentials
-
VolCon is a distributed way of running jobs on the cloud based on docker, without requiring the BOINC client or VirtualBox
-
Although it can run on any system with an unique IP address (such as internal networks accessible from the main BOINC server), it is primarily designed for cloud servers
-
By default, the main BOINC server does not provide VolCon and so any jobs submitted to it will never be run
-
To create a VolCon network, run the following steps (bash commands below): 0. Modify MySQL to account for VolCon jobs
- Assign VolCon credentials: provide VolCon password
- Start VolCon network: In order for VolCon to be more effective, a large set of working nodes is required. Here will show how to manually run a simple mirror container.
- Assign Runner credentials: Multiple types of VolCon runners can be used at the same type (GPUs or not, different providers, etc). All runners within the same provided organization type are assumed to be similar and so they can be requested in such a way if the user wishes to.
- Implement VolCon runners by organization: The example here must be run within a VM, can be run on multiple machines without any changes.
-
Note: VolCon is an update over the previously known ADTD-P system and, although both share many features, VolCon is not backwards compatible
-
# (0) MySQL container, mysql
use boincserver;
CREATE TABLE IF NOT EXISTS volcon_jobs (
job_id BIGINT AUTO_INCREMENT UNIQUE,
Token VARCHAR(255),
Image VARCHAR(255),
Command VARCHAR(5000),
Date_Sub DATETIME,
Date_Run DATETIME,
Error VARCHAR(255),
Notified VARCHAR(255),
status VARCHAR(255),
computation_time DOUBLE,
download_time DOUBLE,
GPU bool,
received_time DATETIME,
mirror_ip VARCHAR(255),
volcon_id VARCHAR(255),
priority VARCHAR(255),
PRIMARY KEY (job_id)
);
# (1) Apache container
python3 /home/boincadm/project/BOINCatTACC/create_VolCon_distributors.py
# (2) Mirror server
git clone https://github.com/ritua2/BOINCatTACC
cd BOINCatTACC/volcon-mirrors
# Build the image, image is not available in dockerhub
docker build -t boinc_tacc/volcon-mirrors:latest .
docker run -d -p 7000:7000 -e "main_server=boinc.tacc.utexas.edu" \
-e "volcon_key=$volcon_key" boinc_tacc/volcon-mirrors:latest
# (3) Apache container
python3 /home/boincadm/project/BOINCatTACC/create_VolCon_clusters.py
# (4) Client server
git clone https://github.com/ritua2/BOINCatTACC
cd BOINCatTACC/volcon-clients
# Build the image, image is not available in dockerhub
docker build -t boinc_tacc/volcon-clients:latest .
# Set GPU=1 to allow GPU jobs, set as 0 as default
docker run -d -p 8000:8000 -e "main_server=boinc.tacc.utexas.edu" \
-e "cluster=andromeda" -e "cluster_key=m110" -e "GPU=1" \
boinc_tacc/volcon-clients:latest
- Disconnecting the APIs, including Reef:
- Disconnecting the APIs will not delete any files currently saved in Reef
cd /home/boincadm/project
./API_Daemon.sh -down
- Pulling the APIs up again:
cd /home/boincadm/project
./API_Daemon.sh -up
-
I just submitted a job and it does not appear in the results ops page
- boinc-docker-server needs to download the images before any further processing is done. For large images (> 1 Gb) this process can take time if the server has a low amount of RAM.
-
I received a job with an empty compressed results file
Try to obtain your results in a regular file, since those are the files that our file retrieval tool supports. In general, use the general outputs or texts files. Avoid images and binary files for results. Using plotting inside a BOINC job may cause problems, since most graphic libraries assume a screen.
The OpenFOAM6 application that users can run through BOINC is licensed under GPL.
This application is built in container form, with the only change being a python script that moves the output results so that the BOINC client may return them to the server
A copy of the original GPLv3 license is provided here, users may also find it in the OpenFOAM6 main Github repository.