BurScale is an autoscaling system which uses a combination of regular and burstable virtual machine instances to save in costs. BurScale is published as a research paper in ACM Symposium on Cloud Computing 2019 (SoCC'19). You can use the following bibTex to cite the paper:
@inproceedings{Baarzi:2019:BUB:3357223.3362706,
author = {Baarzi, Ataollah Fatahi and Zhu, Timothy and Urgaonkar, Bhuvan},
title = {BurScale: Using Burstable Instances for Cost-Effective Autoscaling in the Public Cloud},
booktitle = {Proceedings of the ACM Symposium on Cloud Computing},
series = {SoCC '19},
year = {2019},
isbn = {978-1-4503-6973-2},
location = {Santa Cruz, CA, USA},
pages = {126--138},
numpages = {13},
url = {http://doi.acm.org/10.1145/3357223.3362706},
doi = {10.1145/3357223.3362706},
acmid = {3362706},
publisher = {ACM},
address = {New York, NY, USA},
keywords = {autoscaling, burstable instances, cloud computing, resource provisioning},
}
BurScale is implemented on top of AWS, so all the instructions bellow assume that you're operating on AWS. BurScale relies on two components: loadbalancer configurator and burscale controller:
To deploy BurScale, you have to make sure that your ingress loadbalancer exposes enough endpoints for BurScale such as adding/removing node, updating the dispatching policy, and collecting the statistics.
For example, if you use Nginx as your loadbalancer, you may use loadcat which is a loadbalancer configurator tool for Nginx. We recommend to use our own fork of loadcat which supports different types of instances including burstable instances.
To run the loadcat daemon along with your loadbalancer node, run the binary:
./loadcat &
This will run the loadcat daemon and exposes port 26590 for BurScale usage.
Before deploying the BurScale controller, you have to create/configure a loadbalancer using loadcat. Browse LOADBALABCER-DNS:26590/balancers/new to create and configure the loadbalancer. For more information please refer the loadcats documentation.
After configuring the loadbalancer, your can deploy the burscale controller.
In order to deploy BurScale, you have to provide a configuratoin file which contains necessary configurations for BurScale controller:
"mu": The service rate of the application.
"default_weight": The initial weight that is used for loadbalancer.
"baseline": The baseline CPU utilization for burstable instances.
"init": Initial number of regular instances for the cluster.
"btype": The type of burstable instances that are used.
"regtype": The type of regular instances that are used.
"image_id": The ID of AMI that is used for the application.
"az": The AWS availablity zone that the application is deployed.
"sg": The AWS security group that the application is deployed.
"snid": The AWS subnet ID that the application is deployed.
"keyname": The AWS access key name that the instances are created.
"duration": The window size (in seconds) in which the autoscaler makes descision.
"balancer_id": The balancer ID that loadcat creates.
"balancer_host_name": The DNS for the loadbalancer.
"balancer_label": The loadbalancer's label created by loadcat.
"region_name": AWS region that the application is deployed.
"logfile": BurScale's logging file name.
Once the loadbalancer is created and configured and also the configuratiotion file for BurScale is ready, you can deploy the BurScale controller. You can either use a docker image or directly use the source files.
A Dockerfile is provided for ease of use. Follow these instructions to build, push, pull, and finally run the docker container containing BurScale controller.
Build:
docker build -t <your-username>/burscale:latest .
Push:
docker push <your-username>/burscale:latest
On your loadbalancer node, pull the image:
docker pull <your-username>/burscale:latest
Finally run the container:
docker run -v ./config.json:config.json burscale config.json
If you want to use the source directly, follow the instructions bellow:
pip3 install -r requirements.txt
cd ./burscale
python3 controller.py config.json