Autonomic Vertical Elasticity of Docker Containers with ELASTICDOCKER

Abstract

Elasticity is the key feature of cloud computing to scale computing resources according to application workloads timely. In the literature as well as in industrial products, much attention was given to the elasticity of virtual machines, but much less to the elasticity of containers. However, containers are the new trend for packaging and deploying microservices-based applications. Moreover, most of approaches focus on horizontal elasticity, fewer works address vertical elasticity. In this paper, we propose ELASTICDOCKER, the first system powering vertical elasticity of Docker containers autonomously. Based on the well-known IBM's autonomic computing MAPE-K principles, ELASTICDOCKER scales up and down both CPU and memory assigned to each container according to the application workload. As vertical elasticity is limited to the host machine capacity, ELASTICDOCKER does container live migration when there is no enough resources on the hosting machine. Our experiments show that ELASTICDOCKER helps to reduce expenses for container customers, make better resource utilization for container providers, and improve Quality of Experience for application end-users. In addition, based on the observed migration performance metrics, the experiments reveal a high efficient live migration technique. As compared to horizontal elasticity, ELASTICDOCKER outperforms Kubernetes elasticity by 37.63%.

BARISTA: Efficient and Scalable Serverless Serving System for Deep Learning Prediction Services

“…These prediction services are usually containarized [3] and encapsulated with all the required software packages. Thus, deployer of these services can preferably use the functionas-a-service (FaaS) approach to hosting these services in an event-driven manner, wherein the functions are executed on the occurrence of some trigger or event (e.g., incoming request).…”

“…Vertical elasticity adds flexibility as it eliminates the overhead in starting a new VM and loading the service model. Prior efforts to scale the CPU resources vertically appear in [24], [25] including an approach that uses the discrete-time feedback controller leveraging MAPE-K loop for containerized applications [3]. Barista uses an efficient, proactive method to trigger the scaling of resources horizontally while relying on vertical scaling reactively to allocate and de-allocate CPU cores for model correction when our estimation model cannot predict accurately.…”

“…In [34], a hybrid model called Prophet is proposed for forecasting workloads by combining linear/logistic trend models with a Fourier series-based seasonality model. According to the authors, Prophet is easier to use than the widely used ARIMA models [3], [35] as it handles missing values automatically. Moreover, the ARIMA models generally struggle to produce good quality forecasts as it lacks seasonality detection.…”

Hot-starting software containers for bioinformatics analyses

“…Docker currently integrates CRIU as an experimental checkpoint sub-command that saves the state of processes to a collection of files on disk. The checkpointing command has been used to migrate containers from the source host to target host when the resources of the source are limited 8 , for fault tolerance purposes 9 , and can provide highly available and scalable of micro-services 10 .…”

Coordinating Vertical Elasticity of both Containers and Virtual Machines

“…To experiment this mechanism, we implement live migration technique for Docker containers. CRIU (Checkpoint/Restore, 2017) is used to achieve the procedure and migrate containers lively (Al-Dhuraibi et al, 2017a). CRIU is a Linux functionality that allows to checkpoint/restore processes, e.g., Docker containers.…”